AI Distilled

An Intuitive Intro to RLAI_Distilled #76: Align Meta Llama 3 to human preferences with DPOWelcome to AI_Distilled. Today, we’ll talk about:Awesome AI:Polymet - Idea to prototype within secondsClipAnything - Choppityfal.aiEarkick - Your Personal AI ChatbotOuterbase | The interface for your databaseMasterclass:Voice Trigger System for SiriAlign Meta Llama 3 to human preferences with DPOAn Intuitive Intro to RLEnhancing LLMs with Structured Outputs and Function CallingSafely repairing broken builds with MLHackHub:Agents for software development Open-source LLM app development platformbuild, manage & run useful autonomous agentsUnderstand Human Behavior to Align True NeedsGenerative models for conditional audio generationCheers!Shreyans SinghEditor-in-Chief, Packt💻 Awesome AI: Tools for WorkPolymet - Idea to prototype within secondsPolymet is an AI-powered tool that helps users quickly turn ideas into prototypes by generating designs and production-ready code in seconds. Users can describe what they need, iterate on the design with their team, and then export the code and designs, which can easily integrate with tools like Figma and existing codebases.ClipAnything - ChoppityChoppity is an AI-powered video editing tool that allows users to quickly find and clip moments from any video using visual, audio, and sentiment analysis. With its "ClipAnything" feature, users can search for specific parts of a video, such as key events, people, or emotions, without having to manually review hours of footage.fal.aiFal.ai is a generative media platform designed for developers to create and deploy AI-powered applications, particularly focused on text-to-image models. It offers fast, cost-effective inference with models like FLUX.1 and Stable Diffusion, optimized for various creative tasks.Earkick - Your Personal AI ChatbotEarkick is an AI-powered mental health app that helps users track and improve their emotional well-being in real time through a personal chatbot named Panda. Earkick tracks mental readiness, mood, and calmness, while providing daily insights, breathing techniques, and guided self-care sessions.Outerbase | The interface for your databaseOuterbase is an AI-powered platform that simplifies working with databases for engineers, researchers, and analysts. It supports SQL and NoSQL databases, allowing users to manage data securely while using AI tools to write queries, fix mistakes, and generate charts and visualizations instantly. Outerbase's table editor, dashboards, and data catalog help users organize, analyze, and share insights efficiently.🔛 Masterclass: AI/LLM TutorialsVoice Trigger System for SiriApple's voice trigger system for Siri includes a first-stage low-power detector to identify potential triggers, and a second-stage, high-precision model to confirm the trigger. It also incorporates speaker identification to ensure the device responds only to its primary user. This sophisticated setup addresses challenges like background noise and phonetically similar words while maintaining power efficiency and privacy.Align Meta Llama 3 to human preferences with DPODPO involves fine-tuning a large language model (LLM) based on feedback from human annotators who rate or rank the model's responses according to desired values, such as helpfulness and honesty. SageMaker Studio provides the computational environment to fine-tune the model using Jupyter notebooks with powerful GPU instances, while SageMaker Ground Truth simplifies the process of gathering human feedback by managing workflows for data annotation. Together, they allow you to align the Llama 3 model’s responses with specific organizational values efficiently.An Intuitive Intro to RLReinforcement learning (RL) is a type of machine learning where an agent learns by interacting with its environment, making decisions, and receiving feedback in the form of rewards or penalties. The goal is to maximize cumulative rewards over time. The agent starts with little to no knowledge and improves through trial and error, learning from past experiences. In RL, actions taken by the agent change the state of the environment, and based on the rewards received, the agent adjusts its future actions. A key concept in RL is balancing exploration (trying new things) and exploitation (using known strategies for rewards).Enhancing LLMs with Structured Outputs and Function CallingEnhancing LLMs with structured outputs and function calling improves their ability to provide accurate and useful responses. Structured outputs ensure consistency and clarity by organizing information in a logical format, reducing ambiguity. Function calling allows LLMs to perform specific tasks, such as retrieving real-time data or executing external functions, making them more interactive and versatile. Combined with techniques like Retrieval-Augmented Generation (RAG), which integrates relevant external information into the model’s responses, these enhancements lead to more reliable, accurate, and contextually rich conversations with LLMs.Safely repairing broken builds with MLGoogle's engineers have developed a machine learning model called DIDACT to automatically repair broken code builds by analyzing historical data of build errors and their fixes. This model suggests potential fixes to developers directly within their Integrated Development Environment (IDE). In a controlled experiment, the use of these machine learning-suggested fixes improved productivity by reducing active coding and feedback time, and increasing the number of completed code changes.🚀 HackHub: AI ToolsAll-Hands-AI/OpenHandsOpenHands is an AI-powered platform designed to assist with software development, allowing agents to perform tasks similar to human developers. These agents can modify code, run commands, browse the web, call APIs, and even use resources like StackOverflow. OpenHands is easy to set up using Docker and can be run in various modes, including scriptable or interactive CLI.langgenius/difyDify is an open-source platform for developing AI applications, offering an intuitive interface that integrates workflows, agent capabilities, model management, and observability features. Dify's core features include a visual AI workflow builder, integration with numerous LLMs, agent tools, and a retrieval-augmented generation (RAG) pipeline for document handling.TransformerOptimus/SuperAGISuperAGI is an open-source framework designed for developers to create, manage, and run autonomous AI agents. It allows seamless operation of multiple agents simultaneously and provides tools to extend their capabilities. With features like graphical interfaces, performance telemetry, and integration with multiple vector databases, SuperAGI enables AI agents to efficiently handle tasks, learn from experience, and optimize token usage.lllyasviel/Paints-UNDOPaints-Undo is an open-source project that provides AI models designed to simulate the drawing process in digital art. By inputting a completed image, users can generate a sequence of steps showing how that image might have been created, mimicking the "undo" function in digital painting software.Stability-AI/stable-audio-toolsStable-Audio-Tools is an open-source library for working with audio generation models. It provides tools for training and running models that generate audio, including a Gradio interface for testing. Users can install the library via PyPI, and the repository includes scripts for both training models and performing inference.📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

Cursor AI raises $60M AI_Distilled #65: Google launches new Gemini models ChatGPT for Conversational AI and Chatbots This book covers the fundamentals of ChatGPT, its applications in conversation design, and practical uses in various contexts. The book delves into LangChain, a framework for working with language models, teaching readers about prompt engineering, chatbot memory, vector stores, and response validation. It also explores the creation of ChatGPT-powered chatbots that can interact with custom data sources, and guides readers through building chatbot user interfaces. Get it for $35.99 $24.99 Welcome to AI_Distilled. Today, we’ll talk about: Techwave: Google launches new Gemini models Cursor AI raises $60M Artifacts are now generally available \ Anthropic Salesforce introduces two new AI sales agents System Prompts Release Notes for Claude.ai and Mobile Apps Awesome AI: LM Studio - Discover, download, and run local LLMs Painless Data Extraction and Web Automation Fleak AI Serverless API Builder Listen to Actual Clients' Feedback Theysaid - Conversational AI Surveys Masterclass: Unlocking 7B+ language models in your browser: A deep dive with Google AI Edge's MediaPipe Deploying Attention-Based Vision Transformers to Apple Neural Engine Mistral-NeMo: 4.1x Smaller with Quantized Minitron Connect the Amazon Q Business generative AI coding companion to your GitHub repositories Augmenting recommendation systems with LLMs HackHub: high-performance, multiplayer code editor from the creators of Atom and Tree-sitter. Multi-Platform Package Manager for Stable Diffusion Sharpen your low-resolution pictures with the power of AI upscaling Transform your database into your AI platform Large language model series developed by Qwen team, Alibaba Cloud. Cheers! Shreyans Singh Editor-in-Chief, Packt ⚡ TechWave: AI/GPT News & Analysis Google launches new Gemini models Google has announced updates to its experimental Gemini models, including a smaller, improved variant called Gemini 1.5 Flash-8B and a more powerful version named Gemini 1.5 Pro. These models show significant performance gains in areas like coding and handling complex prompts. The updates aim to gather feedback from developers before a full-scale release, with the models available for free testing via Google AI Studio and the Gemini API. While some praise the rapid improvements, others criticize the models for still struggling with longer tasks and coding reliability. Cursor AI raises $60M AI startup Cursor, founded by four MIT friends, has gained popularity for its AI-powered code completion tools, now used by engineers at top AI companies like OpenAI and Midjourney. Recently, Cursor raised $60 million in a Series A funding round, bringing its valuation to $400 million. The software, built on large language models like GPT-4, helps developers automate tedious coding tasks, making it easier to fix bugs and build prototypes. With over 30,000 users, Cursor aims to revolutionize coding by allowing engineers to focus more on creativity and complex problem-solving. Artifacts are now generally available \ Anthropic Claude has made its Artifacts feature available to all users across Free, Pro, and Team plans, including on iOS and Android apps. Artifacts allow users to create, view, and iterate on various work products, like code snippets, flowcharts, and interactive dashboards, directly within their conversations with Claude. Since its preview launch in June, tens of millions of Artifacts have been created. Salesforce introduces two new AI sales agents Salesforce has introduced two new AI-powered sales agents: Einstein SDR Agent and Einstein Sales Coach Agent, both launching in October. Einstein SDR Agent autonomously manages inbound leads, answering questions, handling objections, and scheduling meetings, freeing up sales teams to focus on more complex tasks. Einstein Sales Coach Agent helps sales representatives improve their skills by simulating buyer interactions and providing feedback. These tools, built on Salesforce’s Einstein 1 Agentforce Platform, aim to enhance sales productivity and effectiveness, with companies like Accenture planning to use them to manage complex deals and scale operations. System Prompts Release Notes for Claude.ai and Mobile Apps Anthropic has introduced a new section in their documentation to log updates to the default system prompts used in conversations on Claude.ai and its mobile apps. These prompts guide how Claude interacts with users, providing up-to-date information and encouraging specific behaviors, like using Markdown for code snippets. The updates to these system prompts aim to improve Claude’s responses but do not affect the Anthropic API. 💻 Awesome AI: Tools for Work LM Studio - Discover, download, and run local LLMs LM Studio 0.3.0 is a major update to the local LLM desktop application that enhances its offline capabilities with new features. Users can now chat with documents, using either full document context or "Retrieval Augmented Generation" (RAG) for longer texts. The update also introduces an OpenAI-like JSON output API, customizable UI themes, and automatic hardware detection for optimal performance. Painless Data Extraction and Web Automation (agentql.com) AgentQL is a powerful tool for data extraction and web automation that uses AI to reliably find and interact with web elements, even as websites change. Unlike traditional methods that rely on fragile XPath or DOM selectors, AgentQL allows users to locate elements using natural language descriptions, making it easier to automate tasks like filling forms, gathering data, and conducting end-to-end testing. Fleak AI Workflows. Simplified | Serverless API Builder | fleak.ai Fleak is a low-code, serverless API builder designed for data teams to quickly and easily create, integrate, and scale AI and data workflows without managing any infrastructure. It allows users to configure and deploy workflows in minutes, seamlessly integrating with tools like large language models, vector databases, and modern storage technologies. Listen to Actual Clients' Feedback | Seven24 AI Seven24 helps you capture and act on user feedback with ease. Integrate their tool into your product to collect feedback via text or voice, and their AI transforms this feedback into actionable tasks. With features like sentiment analysis, you can boost positive reviews and address issues quickly. Theysaid - Conversational AI Surveys TheySaid offers the world’s first conversational AI survey, designed to significantly increase response rates and improve customer engagement. By integrating seamlessly with your existing tech stack, the AI tool generates personalized survey questions based on your website content and follows up with users through conversational interactions. 🔛 Masterclass: AI/LLM Tutorials Unlocking 7B+ language models in your browser: A deep dive with Google AI Edge's MediaPipe Google AI Edge's MediaPipe has developed a new system that allows large language models (LLMs) to run directly in web browsers, overcoming memory and performance limitations. By using WebAssembly and WebGPU, MediaPipe can now load and execute models like Gemma 1.1 with 7 billion parameters, which was previously unfeasible in-browser. The approach includes breaking down models into manageable parts and leveraging efficient memory usage techniques to handle the massive size of LLMs. Deploying Attention-Based Vision Transformers to Apple Neural Engine The concept of Vision Transformers (ViTs) was introduced to leverage transformer models, which were originally used in natural language processing, for image recognition tasks. Unlike traditional Convolutional Neural Networks (CNNs), Vision Transformers process images by dividing them into smaller patches and applying attention mechanisms. This approach can handle various computer vision tasks such as image classification and object detection more effectively. Mistral-NeMo: 4.1x Smaller with Quantized Minitron NVIDIA's Minitron technique makes large language models (LLMs) like Mistral-NeMo smaller and more efficient by removing less critical parts and retraining them. This process reduces the models' sizes while keeping their performance high. The Minitron version of Mistral-NeMo, for instance, shrinks the model from 12 billion to 8 billion parameters. Combining Minitron with 4-bit quantization further compresses these models, allowing them to run on smaller GPUs and reducing operational costs. Connect the Amazon Q Business generative AI coding companion to your GitHub repositories You can link Amazon Q Business, an AI-powered assistant, to your GitHub repositories using the Amazon Q GitHub (Cloud) connector. This setup allows you to use natural language queries to access information like commits, issues, and pull requests from your GitHub repositories. By integrating this tool, your development team can boost productivity, reduce context switching, and quickly retrieve information from your GitHub data through a conversational interface. Augmenting recommendation systems with LLMs Large language models (LLMs), like Google's PaLM, can significantly enhance recommendation systems by integrating advanced AI capabilities. By incorporating LLMs into the recommendation pipeline, you can improve features like conversational recommendations, sequential recommendations based on user activity, and rating predictions. LLMs can interactively suggest items, understand the sequence of user preferences, and predict ratings with high accuracy. 🚀 HackHub: AI Tools zed-industries/zed Zed is a high-performance, multiplayer code editor developed by the team behind Atom and Tree-sitter. It can be installed on macOS and Linux directly or through package managers, though it’s not yet available for Windows or web platforms. LykosAI/StabilityMatrix Stability Matrix is a multi-platform tool designed for managing Stable Diffusion Web UI packages across Windows, Linux, and macOS. It features a customizable interface with a syntax-highlighted terminal, a model browser for importing models from CivitAI and HuggingFace, and a shared model directory for all packages. Lucchetto/SuperImage SuperImage is an Android app that uses AI to enhance low-resolution images by upscaling them to higher resolutions. Built with the MNN framework and Real-ESRGAN, it processes images in tiles on the device's GPU, merging them into a high-resolution final image. It requires Android 7 or above and support for Vulkan or OpenCL. superduper-io/superduper Integrate AI models and machine learning workflows with your database to implement custom AI applications, without moving your data. Including streaming inference, scalable model hosting, training and vector search. QwenLM/Qwen2 Qwen2 is a suite of advanced language models available in various sizes, including up to 72 billion parameters. It offers state-of-the-art performance in tasks like coding and math, and supports up to 128K tokens for extended context. The models are pretrained and instruction-tuned, and they are available for use through Hugging Face and ModelScope. 📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us. If you have any comments or feedback, just reply back to this email. Thanks for reading and have a great day! *{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{line-height:0;font-size:75%} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

Learn all that happened at the AI Action SummitAI_Distilled #82: We are back!AI is not the FUTURE, it’s the PRESENT! Here’s How to NOT Get Left Behind!Want to be ahead of the curve?Block 3 hours of your time to learn AI tools & workflows that 99% of people don’t know yet!🗓️ Tomorrow | ⏱️ 10 AM ESTIn this training, you’ll learn how to:✅ Master 30+ AI tools to automate work & increase efficiency✅ Save 1000s of dollars by leveraging AI for business & personal growth✅ Eliminate repetitive tasks & boost creativity effortlessly✅ Use AI to analyze data, make smarter decisions, and scale fasterHurry! Click here to register (FREE for the first 100 people only!)Hi, there!Greetings for 2025! We’ve been off the radar for a while as we worked on re-inventing our content offerings. AI Distilled will now be run by the LLM Expert Insights team, and we promise to make it up to you with exciting offers in the coming weeks.LLM Expert Insights Team,PacktNewsA two-day AI Action Summit was held in Paris, France on February 10-11, 2025. The summit brought together governments, public and private organizations, academia, NGOs, artists, and civil society. Core themes included public interest AI, the future of work, innovation and culture, trust in AI, and global AI governance. Some of the key announcements were: AI Action Summit Declaration 73 participating members, including 27 EU states, governments, research institutes, and government bodies signed the statement on inclusive and sustainable AI for people and the planet. The UK and the US refrained from signing the declaration. EU launches InvestAI initiative to mobilise €200 billion of investment in artificial intelligence   The InvestAI initiative was announced at the Paris summit with a pledge of EUR 150 billion from the private sector and EUR 50 billion from the public sector. This initiative will support the computing power for the world’s fastest public supercomputers. Ursula von der Leyen, the EU Commission President, vowed in her speech to cut red-tape in AI while ensuring safe AI, encouraging the collaboration of global talent with AI Gigafactories. Launch of public interest initiatives   Current AI, an international partnership of governments, philanthropists, and industry, was officially launched at the AI Action Summit with $400 million in funding, shared Martin Tisné, CEO of AI Collaborative, in his LinkedIn post. Robust Open Online Safety Tools (ROOST)  a non-profit organization incubated at The Institute of Global Politics at Columbia University was also launched at the summit. ROOST has some of the biggest names in the industry as founding partners, including Google, Discord, OpenAI, Roblox and GitHub, Hugging Face, Microsoft, Wikimedia, among others. ROOST aims to provide open-source building blocks and safety resources to global users and communities. OpenAI Roadmap announcedOpen AI will now focus on simplifying product offerings and unify o-series and GPT series models. There will be no o3 release, but GPT-5 will be rolled out with a higher-level intelligence setting for Pro and Plus subscribers and standard intelligence for free tier users.Groq secures $1.5bn from Saudi Arabia to expand AI inference infrastructure in the region Groq CEO Jonathan Ross announced in a LinkedIn post a $1.5 billion agreement to expand Groq’s LPU-based AI infrastructure. This investment will support Groq’s existing data centre in Saudi Arabia and fuel the development of the Arabic Large Language Model (ALLaM).  Elon Musk-Led Group Makes $97.4 Billion Bid for Control of OpenAI, SamA not interested  A group of investors led by Elon Musk has offered to buy control of OpenAI for $97.4 billion. This bid introduces a new twist in OpenAI’s future as the company moves towards restructuring in order to transition to a for-profit entity. The bid backed by xAI, Baron Capital Group, Emanuel Capital Management, 8VC, Valor, Atreides, and Vy Capital is Musk’s latest attempt to make OpenAI open-source and safety-focused, as confirmed by Musk’s attorney, Marc Toberoff. Sam Altman (SamA) took to X to express disinterest in the offer and instead made a counteroffer. 💻 Awesome AI: Tools for WorkMeet New Perplexity Sonar Perplexity has released an optimized version of Sonar to improve decoding throughput which now reaches 1,200 tokens per second. Graded on a scale of 1 to 100, Perplexity’s experiments report that Sonar now scores 85.1 on factuality and 85.9 on readability, surpassing other frontier models. The latest version of Sonar is now available in default search mode for Perplexity Pro users. Cursor’s AI Agent Gets New Capabilities Cursor has added new features to its agent that allow it to accomplish end-to-end development tasks while collaborating with programmers. Some of these features include understanding codebase context, automatically writing and running terminal commands with a programmer’s permission and detecting and fixing lint errors. GitHub Copilot: The agent awakens - The GitHub Blog GitHub unveiled Project Padawan to introduce Copilot’s autonomous agent. In agent mode, Copilot utilizes a SWE agent that can suggest terminal commands, recognize and fix errors, walk through its code, analyse its output and result, debug, diagnose, and fix errors. Apart from this, GitHub also announced the GA of Copilot Edits in VS Code to help developers make inline changes to multiple files in their workspace using natural language. HackhubHugging Face announces AI Energy Score Ratings To drive the adoption of energy-efficient AI, Hugging Face launched the AI Energy Score project. This project offers standardized benchmark for the energy consumption of various AI models. Developers can submit their models to be assessed by a uniform framework and obtain a star rating for their models. There is also a leaderboard that presently ranks 166 models. Go check it out. Open R1 project introduces OpenR1-Math-220k After launching the OpenR1 project to reproduce DeepSeek-R1’s data and training pipeline, the community, in collaboration with Project Numina, announced the construction of OpenR1-Math-220K generated by prompting DeepSeek-R1. Anthropic Economic Index Anthropic analyzed Claude.ai’s anonymized conversations to study how AI is used in real-world tasks and its impact on the labor markets. The study found that 37.2% of conversations were centered around computer and mathematical domains. Computer programmers and copywriters with mid-to-high-median salaries were the highest AI users. The dataset and report have been open sourced. LumaLabsAI drops image to video model In an X post, LumaAI announced the release of image-to-video generation using the Ray2 model. Users subscribed to LITE or PLUS plans can drop any image into the Dream-Machine and create realistic videos. ByteDance introduces OmniHuman-1 ByteDance has released an AI framework that can generate human videos from a single image and motion signal. This diffusion-transformer-based animation framework uses multiple modalities (audio, video, and a combination of signals) to achieve realistic human video generation. TechwaveOpen AI introduces the Intelligence Age with its SuperBowl debut ad To reach the masses, Open AI positioned ChatGPT as the precursor to the Intelligence Age in its first-ever television ad. The ad showcased AI as a tool and brainstorming partner to “assist, aid, and enhance” human-led product vision. Sam Altman’s views on the economics of AI SamA noted in his blog that investing money and resources in AI will drive gains in intelligence for AI models and that the cost of using AI will continue to drop over time, allowing for its wider adoption. He also announced the rollout of AI agents capable of replacing junior level software engineers, potentially impacting jobs and the economy. MasterclassMeta is working on Pippo, a generative model for turnaround videos of humans using a single image Pippo is a multi-view diffusion transformer model pre-trained on 3 billion uncaptioned human images, using both full-reference and cropped versions. It also uses head orientation, position (2D projected anchor), and target camera viewpoint as input.  The model undergoes mid-training on low-resolution images and post-training on high-resolution studio camera images of humans. While the mid-training phase uses an MLP, a ControlNet-inspired MLP is applied to create a 3D-aware multi-view model. Visit here for a visual demo. Decoding-based Regression - Google DeepMind Researchers at DeepMind investigated the use of LLMs to perform regression task by representing numeric predictions as decoded strings and using auto-regressive prediction. They experimented with both normalized and un-normalized tokenization. The proposed approach performed as well as traditional approaches, can be applied to density estimation tasks, and could capture distributions modelled over Gaussian and Riemann distributions. Tell us more about your content needsWe would love to hear from you! Fill out this form to tell us what you’d like to read in AI Distilled next.📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

Replit Agent early accessAI_Distilled #67: Apple Intelligence comes to iPhone, iPad, and Mac starting next monthGrow your business & career by 10x using AI Strategies in 4 hrs! 🤯Imagine a future where your business runs like a well-oiled machine, effortlessly growing and thriving while you focus on what truly matters.This isn't a dream—it's the power of AI, and it's within your reach.Join our AI Business Growth & Strategy Crash Course and discover how to revolutionize your approach to business on 12th September at 10 AM EST.In just 4 hours, you’ll gain the tools, insights, and strategies to not just survive, but dominate your market.Sign up here to save your seat! 👈Welcome to AI_Distilled. Today, we’ll talk about:Techwave:[Sponsored] Grow your career by 10x using AI Strategies in 4 hrs!Apple Intelligence comes to iPhone, iPad, and Mac starting next monthReplit Agent early accessAI system developed by Google DeepMind that designs novel proteinsIntroducing LLaVA V1.5 7B on GroqCloudFunction Calling in Google AI StudioAwesome AI:Polymet - Idea to prototype within secondsClipAnything - Choppityfal.aiEarkick - Your Personal AI ChatbotOuterbase | The interface for your databaseMasterclass:Voice Trigger System for SiriAlign Meta Llama 3 to human preferences with DPOAn Intuitive Intro to RLEnhancing LLMs with Structured Outputs and Function CallingSafely repairing broken builds with MLHackHub:Agents for software development Open-source LLM app development platformbuild, manage & run useful autonomous agentsUnderstand Human Behavior to Align True NeedsGenerative models for conditional audio generationCheers!Shreyans SinghEditor-in-Chief, Packt💡Recommended Reading: Essential Concepts of Vector DatabasesUnderstand why vector databases are important in modern data management and how to use them effectively.The course is about 4 hours long and is aimed at people interested in advanced data management techniques.The course includes hands-on sessions for setting up and using these databases, as well as integrating them with Large Language Models and frameworks like LangChain.Get it for $84.99⚡ TechWave: AI/GPT News & AnalysisApple Intelligence comes to iPhone, iPad, and Mac starting next monthApple announced the launch of "Apple Intelligence," a personal intelligence system integrated with iOS 18, iPadOS 18, and macOS Sequoia, starting in October 2024. This system uses advanced generative models and personal context to enhance everyday tasks, like writing assistance, smarter notifications, and a more flexible Siri. Features like a photo Clean Up tool, transcription in Notes and Phone apps, and AI-powered email prioritization will debut first in the U.S., with expanded language and feature support in the following months.Replit Agent early accessReplit Agent is an AI tool that helps users create software projects by understanding natural language prompts. Currently in early access for Replit Core and Teams subscribers, it assists in building web-based applications by guiding users through each step, from selecting technologies to deploying the final product. The agent is designed for prototyping and works closely with users to refine and develop their applications.AI system developed by Google DeepMind that designs novel proteinsAlphaProteo is an AI system developed by Google DeepMind that designs novel proteins to bind to specific target molecules. This technology can accelerate biological research by creating protein binders that aid in drug development, disease understanding, and more. AlphaProteo builds on the success of AlphaFold but goes further by generating new proteins, not just predicting their structures. It has shown high success rates in binding to key targets, such as proteins involved in cancer and viral infections like SARS-CoV-2.Introducing LLaVA V1.5 7B on GroqCloudLLaVA v1.5 7B is a new multimodal AI model available on GroqCloud, enabling developers and businesses to create applications that integrate image, audio, and text inputs. Built from a combination of OpenAI’s CLIP and Meta’s Llama 2, LLaVA v1.5 excels in tasks like visual question answering, image captioning, and multimodal dialogue.Function Calling in Google AI StudioGoogle AI Studio now supports function calling, allowing users to easily test the model's capabilities directly in the interface. This new feature makes it more convenient to experiment with the AI without leaving the UI. Google AI Studio offers free fine-tuning.💻 Awesome AI: Tools for WorkPolymet - Idea to prototype within secondsPolymet is an AI-powered tool that helps users quickly turn ideas into prototypes by generating designs and production-ready code in seconds. Users can describe what they need, iterate on the design with their team, and then export the code and designs, which can easily integrate with tools like Figma and existing codebases.ClipAnything - ChoppityChoppity is an AI-powered video editing tool that allows users to quickly find and clip moments from any video using visual, audio, and sentiment analysis. With its "ClipAnything" feature, users can search for specific parts of a video, such as key events, people, or emotions, without having to manually review hours of footage.fal.aiFal.ai is a generative media platform designed for developers to create and deploy AI-powered applications, particularly focused on text-to-image models. It offers fast, cost-effective inference with models like FLUX.1 and Stable Diffusion, optimized for various creative tasks.Earkick - Your Personal AI ChatbotEarkick is an AI-powered mental health app that helps users track and improve their emotional well-being in real time through a personal chatbot named Panda. Earkick tracks mental readiness, mood, and calmness, while providing daily insights, breathing techniques, and guided self-care sessions.Outerbase | The interface for your databaseOuterbase is an AI-powered platform that simplifies working with databases for engineers, researchers, and analysts. It supports SQL and NoSQL databases, allowing users to manage data securely while using AI tools to write queries, fix mistakes, and generate charts and visualizations instantly. Outerbase's table editor, dashboards, and data catalog help users organize, analyze, and share insights efficiently.🔛 Masterclass: AI/LLM TutorialsVoice Trigger System for SiriApple's voice trigger system for Siri includes a first-stage low-power detector to identify potential triggers, and a second-stage, high-precision model to confirm the trigger. It also incorporates speaker identification to ensure the device responds only to its primary user. This sophisticated setup addresses challenges like background noise and phonetically similar words while maintaining power efficiency and privacy.Align Meta Llama 3 to human preferences with DPODPO involves fine-tuning a large language model (LLM) based on feedback from human annotators who rate or rank the model's responses according to desired values, such as helpfulness and honesty. SageMaker Studio provides the computational environment to fine-tune the model using Jupyter notebooks with powerful GPU instances, while SageMaker Ground Truth simplifies the process of gathering human feedback by managing workflows for data annotation. Together, they allow you to align the Llama 3 model’s responses with specific organizational values efficiently.An Intuitive Intro to RLReinforcement learning (RL) is a type of machine learning where an agent learns by interacting with its environment, making decisions, and receiving feedback in the form of rewards or penalties. The goal is to maximize cumulative rewards over time. The agent starts with little to no knowledge and improves through trial and error, learning from past experiences. In RL, actions taken by the agent change the state of the environment, and based on the rewards received, the agent adjusts its future actions. A key concept in RL is balancing exploration (trying new things) and exploitation (using known strategies for rewards).Enhancing LLMs with Structured Outputs and Function CallingEnhancing LLMs with structured outputs and function calling improves their ability to provide accurate and useful responses. Structured outputs ensure consistency and clarity by organizing information in a logical format, reducing ambiguity. Function calling allows LLMs to perform specific tasks, such as retrieving real-time data or executing external functions, making them more interactive and versatile. Combined with techniques like Retrieval-Augmented Generation (RAG), which integrates relevant external information into the model’s responses, these enhancements lead to more reliable, accurate, and contextually rich conversations with LLMs.Safely repairing broken builds with MLGoogle's engineers have developed a machine learning model called DIDACT to automatically repair broken code builds by analyzing historical data of build errors and their fixes. This model suggests potential fixes to developers directly within their Integrated Development Environment (IDE). In a controlled experiment, the use of these machine learning-suggested fixes improved productivity by reducing active coding and feedback time, and increasing the number of completed code changes.🚀 HackHub: AI ToolsAll-Hands-AI/OpenHandsOpenHands is an AI-powered platform designed to assist with software development, allowing agents to perform tasks similar to human developers. These agents can modify code, run commands, browse the web, call APIs, and even use resources like StackOverflow. OpenHands is easy to set up using Docker and can be run in various modes, including scriptable or interactive CLI.langgenius/difyDify is an open-source platform for developing AI applications, offering an intuitive interface that integrates workflows, agent capabilities, model management, and observability features. Dify's core features include a visual AI workflow builder, integration with numerous LLMs, agent tools, and a retrieval-augmented generation (RAG) pipeline for document handling.TransformerOptimus/SuperAGISuperAGI is an open-source framework designed for developers to create, manage, and run autonomous AI agents. It allows seamless operation of multiple agents simultaneously and provides tools to extend their capabilities. With features like graphical interfaces, performance telemetry, and integration with multiple vector databases, SuperAGI enables AI agents to efficiently handle tasks, learn from experience, and optimize token usage.lllyasviel/Paints-UNDOPaints-Undo is an open-source project that provides AI models designed to simulate the drawing process in digital art. By inputting a completed image, users can generate a sequence of steps showing how that image might have been created, mimicking the "undo" function in digital painting software.Stability-AI/stable-audio-toolsStable-Audio-Tools is an open-source library for working with audio generation models. It provides tools for training and running models that generate audio, including a Gradio interface for testing. Users can install the library via PyPI, and the repository includes scripts for both training models and performing inference.📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{line-height:0;font-size:75%} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

Perplexity makes DeepResearch free to use, MoonShot AI’s MoBA for long context LLMs, DeepSeek’s Code AI_Distilled #82: We are back! CLICK HERE TO REGISTER Get Early Access - 40% Discount Use Code AGENT40 at checkout This week saw many breakthroughs and announcements and we bring them all together in one place. Our goal is to curate the most relevant news and updates for you. Fill out this form and tell us what you’d like to read next on AI Distilled. LLM Expert Insights Team, Packt 📰 News Google introduces AI co-scientist The AI co-scientist by Google is a multi-agent AI system that is intended to function as a collaborative tool for scientists. It is built on Gemini 2.0 and is designed to mirror the reasoning process underpinning the scientific method. The AI co-scientist can be used to generate novel research hypotheses, a detailed research overview, and experimental protocols. Microsoft unveils Majorana 1 Microsoft has developed a the topoconductor that allows them to create topological qubits and engineer a new state of matter. The toplogical qubits are more stable than traditional qubits, making them more suitable for building a large-scale quantum computer. Microsoft is now gearing towards the next step - building a fault-tolerant quantum computer using topological qubits. Thinking Machines Lab launched, Mira Murati CEO Former OpenAI CTO Mira Murati has launched a startup with Barret Zoph (CTO), John Schulman (Chief Scientist). Various other AI stalwarts who have experience in creating AI products like ChatGPT, Segment Anything, Mistral, Pytorch, Character.ai, OpenAI Gym, and FairSeq are also a part of Thinking Machines Lab. The startup’s core mission is to build intelligent, adaptable, and personalized AI systems, emphasizing human-AI collaboration and safety. It aims to make AI more capable, customizable, understood, and user-friendly. Perplexity Deep Research launched and is free to usePerplexity recently launched its Deep Research model, designed to generate comprehensive reports, using capabilities like iterative search, reasoning, coding, and refinement of research plans. On the Humanity’s Last Exam benchmark test, Perplexity ranked second—behind OpenAI’s deep research model but ahead of other leading competitors—completing most research tasks in under three minutes. Google aims to serve 10 cities with Waymo self-driving cars in 2025 Speaking at the 2025 World Government Summit in Dubai, Google and Alphabet CEO Sundar Pichai talked about expanding Waymo to 10 new cities. He also highlighted Google’s recent achievement in quantum computing and indicated that quantum computers could become mainstream in the next 5 to 10 years. Isomorphic Labs and Novartis expand collaboration Google DeepMind partner Isomorphic Labs, an AI-first drug discovery company, and Novartis have extended their collaboration to add three more research programs aimed at accelerating drug discovery research. Isomorphic Labs is augmenting the AlphaFold breakthrough to connect research with biotech, drug discovery, and medical design. HP acquires Humane’ AI capabilities including the AI platform Cosmos; end of the road for AI Pins HP is acquiring Humane in a $116 million deal to accelerate the development of an intelligent ecosystem across its products and services. Humane has also announced the end of production and consumer availability. AI Pin’s services, features, and data access will be available till February 28, 2025, 12 pm PST. Grok 3 launched; Musk claims it is the Smartest AI Grok 3, a chatbot built in less than a year, was launched this week in a live demo by the xAI team. The live demonstration showcased Grok 3 handling tasks such as creating a launch plan from Earth to Mars and back and an “insanely great game”, a hybrid between Tetris and Bejeweled. The team claimed that Grok 3’s SOTA model is better than DeepSeek, Claude, and Gemini and is comparable to OpenAI’s model. Check out the recorded demo here (at 19:11 seconds). Project Waterworth, a subsea cable connectivity project by Meta Meta has announced a multi-billion-dollar, multi-year project to open three oceanic corridors connecting five major continents. This will be the longest subsea cable project, spanning 50000 kilometers and linking the U.S., Brazil, South Africa, India, and other key regions. Apart from economic collaboration and digital inclusion, this project aims to drive AI innovation across the world with high-speed connectivity. 💻 Awesome AI: Tools for Work Moonshot AI introduces MoBA that combines Mixture of Experts with sparse attention Following the release of Kimi, Moonshot AI introduced the Mixture of Block Attention (MoBA) model, designed to tackle long conversations and large text. After dividing the text into blocks, MoBA uses a gating mechanism that switches between full and sparse attention, focusing on the most informative blocks, thus reducing computation time. MoBA has been able to maintain competitive performance with 1-million-token context length. Perplexity open-sources DeepSeek R1776 to mitigate bias and censorship To tackle DeepSeek’s avoidance of censored topics in China, Perplexity compiled a dataset of 40k multilingual prompts covering 300 censored topics. R1 was then post-trained on this censorship dataset using an adapted NeMo 2.0 Nvidia framework. The model weights can be downloaded from Hugging Face. Mistral Saba, a custom-trained model for Middle East and South Asian regional languages Mistral has introduced Saba, which has been trained on datasets curated from South Asia and Middle East, to capture cultural and linguistic nuances whilst providing accurate and relevant responses to cater to customers in these regions. Meta Segment Anything Model (SAM) 2.1 is now available in Amazon SageMaker Jumpstart The SOTA vision segmentation model, SAM 2.1, is now publicly available through Amazon SageMaker Jumpstart. SAM 2.1 enables zero-shot object segmentation, object detection using prompts, long-context processing, and context segmentation scenarios. 🛠️ Hackhub Hugging Face introduces agent ratings To evaluate the performance of AI agents in real-world business scenarios, Hugging Face has introduced the AI Agent Leaderboard. The leaderboard currently ranks 17 LLMs, evaluated using the Tool Selection Quality (TSQ) metric across 14 multi-domain datasets. This benchmark assesses LLMs on their ability to select appropriate tools for a given query. This includes parameter handling, multi-step decision making, error handling, context management, and reasoning. At present, gemini-2.0-flash-001 is topping the charts with the highest TSQ of 0.938. LlamaIndex introduces LLM Consortium LlamaIndex has introduced a vision for the AI boardroom of the future by creating an LLM consortium, where multiple LLMs answer the same question, and their responses are synthesized by an arbiter to produce a final result. The arbiter iterates and asks the LLMs to try again if it finds their responses subpar. You can check out the notebook here. Meta achieves breakthroughs in decoding language from brain Meta AI can now decode up to 80% of the characters in a sentence using non-invasive brain recordings. Brain2Qwerty, a deep-learning architecture trained on EEG and MEG data, can decode briefly memorized sentences that participants typed on a QWERTY keyboard. In another related experiment, MEG and EEG data was analyzed to capture the neural dynamics of language production in the human brain. ⚙️Techhub Engine AI’s PM01 Robot deployed for public service in Shenzhen 70 Engine AI’s open-source robots are now serving as community workers and patrolling the streets of Shenzhen, in South China. Powered by DeepSeek, the PM01 robot has achieved human-like mobility and is now making grassroot governance more efficient. YouTube integrates Veo2 to Shorts YouTube Shorts is now integrating Google DeepMind’s popular video generation model. Users in the US, Canada, Australia, and New Zealand can now use text prompts in Shorts to generate standalone video footage. Goku AI ByteDance has recently released GokuAI, a generative flow-based image and video generation model trained on millions of image-text and video-text pairs. Built on a transformed based architecture with 1, 2, and 8 billion parameters, Goku uses diffusion techniques, Rectified Flow, and Variational Autoencoder to create high quality visuals that enable business and content creators to amplify their creative applications. 🧠Masterclass DeepSeek researchers introduce CodeI/O, a new technique to improve LLM reasoning DeepSeek researchers recently shared an approach that uses the structured nature of code to learn symbolic, logical, mathematical, and commonsense reasoning patterns. By collecting Python code from sources like CodeMix and PyEdu-R, the code files are unified using DeepSeek-V2.5. The dataset includes 3.5 million input-output pairs generated from transformed code functions, along with natural language Chain-of-Thought (CoT) explanations. During training, DeepSeek is prompted to generate an output (response), with incorrect responses and feedback fed back into the LLM. Instruction tuning is then applied in the second stage. This multi-turn revision enhances accuracy and shows improvements over baseline models. Less is More for Reasoning (LIMO) improves LLM performance with only 1% training data The LIMO approach challenges the notion that LLMs require extensive data, achieving competitive results with just 817 samples and cognitive templates. LIMO employs a rigorous selection process that includes structural organization, effective cognitive explanations, and verification to curate high-quality math problems from NuminaMath-CoT, AIME, and MATH datasets. Using the Qwen2.5-32B-Instruct model with a 16,384-token sequence length, LIMO applies SFT for training and utilizes step-by-step prompting to achieve generalization capabilities. Large Memory Model (LM2) an auxiliary memory-based model for long context reasoning LM2 incorporates a structured memory system that interacts with input embeddings through cross-attention. Built on a decoder-only transformer architecture, the model utilizes memory updates regulated by gating mechanisms, allowing it to selectively retain relevant information. LM2 was tested on the BABILong and MMLU datasets, demonstrating significant improvements in long-context reasoning and general reasoning capabilities. 📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us. If you have any comments or feedback, just reply back to this email. Thanks for reading and have a great day! *{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

xAI, Elon Musk's AI startup, launches an API AI_Distilled #73: Introducing the new Claude 3.5 Sonnet, and Claude 3.5 Haiku and “Computer Use” 🚀 The Most Awaited 2-for-1 Deal Drops Tomorrow! 🚀 Unlock our 2-for-1 offer at Generative AI in Action (Nov 11-13) and bring a friend, colleague, or your team to double the learning experience. 🗓 Sale Starts: Tomorrow, Friday, Oct 25, 10 AM ET ⏳ Duration: 24 hours only Don’t miss out—mark your calendar and get ready to grab this exclusive deal! Join 25+ AI Experts, 30+ Sessions & 1000+ Tech Pros Welcome to AI_Distilled. Today, we’ll talk about: Techwave: xAI, Elon Musk's AI startup, launches an API Introducing Stable Diffusion 3.5 Introducing the new Claude 3.5 Sonnet, and Claude 3.5 Haiku and “Computer Use” Meta releases Spirit LM, open-source multimodal modelintegrating text and speech seamlessly New autonomous agents scale your team like never before Awesome AI: guidde・Magically create video documentation with AI Feta - Better stand-ups, retros, syncs and more BrowserCopilot AI - Your AI Companion Across the Web MyLens.ai: Key Points of any Webpage & Youtube with one click Trag: Superlinter for any stack Masterclass: Solving complex problems with OpenAI o1 models Thinking LLMs:General Instruction Following with Thought Generation Agent-as-a-Judge: Evaluate Agents with Agents Learn dynamic few-shot prompting with LlamaIndexworkflows for enhanced LLM performance Fine-tuning LLMs to 1.58-bit: compress models without sacrificing performance HackHub 3b1b/videos: Code for the manim-generated scenes used in 3blue1brown videos phidatahq/phidata: Build AI Agents with memory, knowledge, tools and reasoning. Chat with them using a beautiful Agent UI. ComposioHQ/composio: Composio equip's your AI agents & LLMs with 100+ high-quality integrations via function calling Janus: Any-to-Anyautoregressive frameworkfor multimodal AI. Ichigo: Llama learns to talk - Homebrew Cheers! Shreyans Singh Editor-in-Chief, Packt ⚡ TechWave: AI/GPT News & Analysis xAI, Elon Musk's AI startup, launches an API Elon Musk’s AI startup, xAI, has launched an API for its generative AI model, Grok, allowing developers to integrate Grok’s features into their applications. The API currently offers a single model, "grok-beta," priced at $5 per million input tokens and $15 per million output tokens. Grok, which powers various features on X (formerly Twitter), is known for its rebellious, uncensored responses and image generation capabilities. Although still developing, xAI aims to catch up to competitors like OpenAI and Anthropic, using data from Musk's companies and X to train future models. Introducing Stable Diffusion 3.5 Stable Diffusion 3.5 is the latest release from Stability AI, offering multiple highly customizable models designed to run on consumer hardware. These models, including Stable Diffusion 3.5 Large and Large Turbo, are available for free for most uses under a permissive license. They offer a balance of high image quality, fast performance, and flexibility, making them ideal for creators, researchers, and businesses. The models can generate diverse images in various styles and are available for download on platforms like Hugging Face. Introducing the new Claude 3.5 Sonnet, and Claude 3.5 Haiku and “Computer Use” Anthropic has announced updates to its Claude 3.5 models, including the upgraded Claude 3.5 Sonnet, which excels in coding and tool use, and the new Claude 3.5 Haiku, which offers similar performance to previous top-tier models at a lower cost and faster speed. They’ve also introduced a groundbreaking “computer use” capability in public beta, allowing Claude to interact with computers like a human by navigating interfaces, clicking buttons, and typing. This feature is still experimental but has potential for automating complex tasks. Meta releases Spirit LM, open-source multimodal modelintegrating text and speech seamlessly Meta has released Spirit LM, a model for handling both spoken and written language in an interleaved manner. The repository contains model weights, inference code, and evaluation scripts for the Spirit LM model, which can be set up using Conda or pip. It includes tools for speech tokenization and text generation, with an emphasis on preserving speech-text sentiment in its outputs. New autonomous agents scale your team like never before Microsoft announced new autonomous agent capabilities in Copilot Studio to help businesses scale more efficiently. Starting next month, businesses will be able to create their own agents, designed to handle tasks like sales, supply chain management, and customer service. These agents, integrated into Dynamics 365, can automate complex processes such as lead generation, supplier communication, and customer support. 💻 Awesome AI: Tools for Work guidde・Magically create video documentation with AI Guidde is an AI-powered platform designed to help businesses quickly create video documentation, making complex workflows easier to explain. It enables users to capture processes using a browser extension or desktop app and automatically generates step-by-step instructions with customizable AI-generated voiceovers. Feta - Better stand-ups, retros, syncs and more Feta is a platform designed to help product and engineering teams run more efficient meetings by streamlining tasks and capturing key insights. It auto-compiles updates for standups, integrates with tools like Jira and GitHub, and generates actionable meeting summaries and notes. BrowserCopilot AI - Your AI Companion Across the Web Yaseen AI is a browser-based AI companion that helps professionals work more efficiently by providing real-time assistance on any website. It integrates seamlessly with workflows, offering personalized responses and support through its Copilot feature. MyLens.ai: Key Points of any Webpage & Youtube with one click MyLens.ai is a Chrome extension that transforms any webpage or YouTube video into visual summaries like mindmaps, timelines, tables, and flowcharts with just one click. It helps users quickly extract key insights from long articles, reports, or videos, saving time by breaking down complex content into clear, shareable visuals. Trag: Superlinter for any stack Superlinter, powered by Trag, is a versatile tool that allows developers to replace traditional linters and code analysis tools with a natural language-based linter that works for any programming language. Users can describe specific code patterns or rules in plain English, which the linter then enforces within their code. 🔛 Masterclass: AI/LLM Tutorials Solving complex problems with OpenAI o1 models Thinking LLMs:General Instruction Following with Thought Generation Large Language Models are typically trained to respond to user instructions based on patterns in data, but they lack the ability to think explicitly before answering. This is important for complex tasks that require reasoning or planning. To address this, a method called Thought Preference Optimization (TPO) allows LLMs to develop thinking abilities without additional human data. The process involves generating multiple potential thoughts, evaluating the quality of the final responses, and optimizing them through reinforcement learning. Agent-as-a-Judge: Evaluate Agents with Agents The "Agent-as-a-Judge" framework is a new method for evaluating agentic systems, where agents are used to evaluate other agents instead of relying on human evaluators or traditional methods that only consider final outcomes. This framework provides feedback throughout the task-solving process, which is important for agentic systems that act step-by-step, like humans. Applied to code generation, "Agent-as-a-Judge" proved more effective and reliable than the existing LLM-as-a-Judge framework and performed similarly to human evaluators, but at a much lower cost and time. Learn dynamic few-shot prompting with LlamaIndexworkflows for enhanced LLM performance In LlamaIndex, workflows are event-driven systems where functions are chained together as steps, each handling specific event types. By using the `@step` decorator, the system ensures that steps only run when a valid event is received, and each step can emit new events for the next. Workflows enable creating processes like agents, document extraction, or retrieval-augmented generation (RAG) pipelines. They are fully asynchronous, allowing efficient parallel processing, and come with built-in observability. Users can integrate global contexts, handle multiple events, and even retry steps in case of failures. Fine-tuning LLMs to 1.58-bit: compress models without sacrificing performance Fine-tuning large language models (LLMs) to use only 1.58 bits per parameter (based on the BitNet architecture) dramatically reduces their computational and memory requirements by using extreme quantization. This process limits the values of each parameter to just three options: -1, 0, and 1. Although such quantization typically requires training a model from scratch, the authors have found ways to fine-tune pre-trained models to achieve similar efficiency without losing significant performance. 🚀 HackHub: AI Tools 3b1b/videos: Code for the manim-generated scenes used in 3blue1brown videos This project contains the code used to create the math videos by 3Blue1Brown, primarily using the Manim library, a tool for generating mathematical animations. While the Manim library itself is open source under the MIT license, the content in this repository is under a Creative Commons license (CC BY-NC-SA 4.0), which allows sharing and adapting with credit but not for commercial purposes. phidatahq/phidata: Build AI Agents with memory, knowledge, tools and reasoning. Chat with them using a beautiful Agent UI. Phidata is a framework for building intelligent agents equipped with memory, knowledge, tools, and reasoning capabilities. You can create agents for various tasks, like web search or financial data analysis, and even combine them into teams to work together. ComposioHQ/composio: Composio equip's your AI agents & LLMs with 100+ high-quality integrations via function calling Composio is a toolset that helps developers build AI agents equipped with a wide range of pre-configured tools and integrations with minimal effort. It simplifies tasks like authentication, accuracy, and reliability, enabling developers to create agents that can interact with platforms like GitHub, Notion, Slack, and more. Janus: Any-to-Anyautoregressive frameworkfor multimodal AI. Janus is an advanced multimodal framework that improves the way AI models understand and generate both visual and textual content. It separates the visual encoding process into distinct pathways but maintains a unified transformer architecture, which increases flexibility and performance for various tasks. Ichigo: Llama learns to talk - Homebrew Ichigo is a new speech and text multimodal model built on Llama3-s, designed for understanding and generating both audio and text. Developed through open research by the Homebrew Computer Company, Ichigo addresses key limitations in earlier models, such as limited multilingual capabilities and issues with recognizing nonspeech inputs. 📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us. If you have any comments or feedback, just reply back to this email. Thanks for reading and have a great day! *{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{line-height:0;font-size:75%} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

OpenAI to become for-profit companyAI_Distilled #69: OpenAI CTO resignsGrow, Make a Difference, and Win!Participate in the Latest Developer Nation Survey!TAKE THE SURVEYWelcome to AI_Distilled. Today, we’ll talk about:Techwave:OpenAI CTO resignsOpenAI to become for-profit companyOpenAI rolls out Advanced Voice ModeSuperintelligence may be here sooner than expected- Sam AltmanEA Unveils Text-to-Game AIAwesome AI:Requstory: convert project ideas into actionable user stories and process maps.Adobe GenStudio: create, manage, and optimize on-brand contentLetta: enhances LLMs by adding memory capabilitiesScenery: AI-powered video editing for teamsKLING AI: Next-Generation AI Creative StudioMasterclass:Vector Embeddings with Cohere and Hugging FaceBuild a multimodal social media content generator using Amazon BedrockWorking with Embeddings: Closed versus Open SourceLinguistic Bias in ChatGPTUpdated production-ready Gemini models, reduced 1.5 Pro pricing, increased rate limits, and moreHackHub:OpenHands: Code Less, Make Moreaudiocraft: library for audio processing and generation with deep learningMidJourney-Styles-and-Keywords-Referencejepa: PyTorch code and models for V-JEPA self-supervised learning from videochat-with-mlx: An all-in-one LLMs Chat UI for Apple Silicon Mac using MLX Framework💡Recommended Reading: LLM Engineer's HandbookCheers!Shreyans SinghEditor-in-Chief, Packt3 Days. 25+ AI Experts. 30+ Sessions.On November 11, join Vin Vashishta, Denis Rothman, John Thompson, Andreas Welsch, and over 20 AI leaders revolutionizing GenAI across industries. From GenAI tools and AI Agents to Small Language Models and LLM fine-tuning, you’ll dive deep into cutting-edge AI strategies and technologies at Packt's Generative AI In Action conference.Don't delay—secure your spot at the early bird rate before prices increase permanently next week!BOOK NOW⚡ TechWave: AI/GPT News & AnalysisOpenAI CTO resignsMira Murati, the Chief Technology Officer of OpenAI, announced her resignation to pursue personal exploration after being with the company for over six years. Murati played a key role in OpenAI's rise, including leading the organization temporarily during a leadership crisis involving CEO Sam Altman. Her departure follows a series of leadership changes at OpenAI, including the exits of other top executives.OpenAI to become for-profit companyOpenAI is planning to restructure into a for-profit benefit corporation, removing control from its non-profit board to make the company more attractive to investors. The non-profit will still exist and hold a minority stake in the for-profit entity. CEO Sam Altman, who has never had equity in OpenAI, will receive equity in the new structure, which could value the company at $150 billion. The move aims to lift investment return caps and make OpenAI more like a typical startup, though it raises concerns about whether the company will maintain its focus on AI safety.OpenAI rolls out Advanced Voice ModeOpenAI has introduced Advanced Voice Mode (AVM) to more ChatGPT users, specifically those in the Plus and Teams tiers, with Enterprise and Edu customers gaining access soon. The new feature enhances ChatGPT's voice interactions, making it more natural to speak with, and includes a redesigned look represented by a blue animated sphere. Users can now choose from five new nature-inspired voices, adding to the existing options.Superintelligence may be here sooner than expected- Sam AltmanOpenAI CEO Sam Altman predicts that superintelligent AI could emerge within the next decade, potentially in "a few thousand days." In a blog post titled "The Intelligence Age," Altman outlines a future where AI accelerates human progress and prosperity, with AI assistants transforming various industries like healthcare and education. He credits deep learning as a key driver of this progress but acknowledges challenges, including labor market disruptions. Altman remains optimistic about AI’s potential to improve lives, urging careful navigation of its risks while aiming for widespread benefits from AI technology.EA Unveils Text-to-Game AIElectronic Arts (EA) unveiled its "Imagination to Creation" vision, allowing players to create video game worlds using simple natural language prompts without coding skills. During a demo, players transformed basic objects into complex, multi-level game environments in real time, using EA's vast library of 3D assets and data. This AI-driven system empowers users to easily generate unique characters, obstacles, and gameplay mechanics.💻 Awesome AI: Tools for WorkRequstory: convert project ideas into actionable user stories and process maps.By simply describing project requirements in natural language, users can generate detailed user stories and visual process maps automatically. The platform allows for easy collaboration, editing, and sharing of these AI-generated documents, streamlining project planning and execution.Adobe GenStudio: create, manage, and optimize on-brand contentAdobe GenStudio is a generative AI-powered tool designed to help marketing teams create, manage, and optimize on-brand content across multiple channels quickly. It provides marketers with AI-driven tools to generate assets, create content variations, and measure performance in real-time, ensuring all content aligns with brand guidelines.Letta: enhances LLMs by adding memory capabilitiesBuilt from research behind MemGPT, Letta helps developers create intelligent agents that can remember and reason over time. It offers tools for building, deploying, and managing AI agents at scale, focusing on memory management and providing a transparent, customizable environment.Scenery video editor | AI-powered video editing for teamsScenery allows users to quickly create and fine-tune videos through a cloud-based system. It simplifies the video editing process with AI-driven tools, such as automatic subject detection, filler word removal, and subtitle generation in over 20 languages. Scenery also enables users to create viral social media clips from longer videos with just a click.KLING AI: Next-Generation AI Creative Studio🔛 Masterclass: AI/LLM TutorialsVector Embeddings with Cohere and Hugging FaceVector embeddings are numerical representations of complex data, like text or images, which help AI models understand and process this data more easily. These embeddings convert input data into dense vectors, where similar data points are close together in a high-dimensional space. This allows AI systems to measure similarities between data points, perform searches, or generate content. Platforms like Cohere and Hugging Face offer pre-trained models that generate embeddings for tasks such as classification, search, and content generation.Build a multimodal social media content generator using Amazon BedrockA multimodal social media content generator using Amazon Bedrock allows brands and content creators to quickly produce visually and textually rich social media posts. The process involves uploading a product image, providing a natural language prompt, and using Amazon Titan Image Generator to create enhanced images. The text for the post is generated using Claude 3, ensuring brand consistency. The system retrieves similar historical posts using Amazon Titan Multimodal Embeddings stored in OpenSearch Serverless, offering suggestions to refine the contentWorking with Embeddings: Closed versus Open SourceEmbeddings are essential in natural language processing (NLP) for tasks like semantic search in retrieval systems. This article explores how different embedding models, both open-source and closed-source, perform in semantic search applications. It discusses techniques like clustering and re-ranking to enhance search results, while comparing the performance, size, and cost of up to nine top models. This comparison helps understand how model size affects efficiency in search tasks, especially when balancing cost and accuracy in large-scale retrieval systems.Linguistic Bias in ChatGPTChatGPT exhibits bias against non-"standard" varieties of English, such as African-American, Indian, and Nigerian English, reinforcing linguistic discrimination. A study comparing responses to different English varieties found that ChatGPT performs worse in understanding, warmth, and naturalness for non-standard varieties, often producing condescending or stereotypical content. While the model imitates some non-standard varieties, it defaults to Standard American English, frustrating non-American users. Even improvements in newer versions like GPT-4 do not fully resolve these issues and, in some cases, worsen stereotyping, highlighting the need for addressing bias in AI.Updated production-ready Gemini models, reduced 1.5 Pro pricing, increased rate limits, and moreGoogle has released updated Gemini models, Gemini-1.5-Pro-002 and Gemini-1.5-Flash-002, with improved performance, lower costs, and faster outputs. These models offer enhanced capabilities for tasks like processing large PDFs, complex math problems, and video analysis. The updates include price reductions of over 50%, higher rate limits, faster output speeds, and reduced latency. The models are designed for general performance across text, code, and multimodal tasks and are available via Google AI Studio and Vertex AI for larger organizations. These updates aim to make the models more efficient and accessible for developers.🚀 HackHub: AI ToolsOpenHands: Code Less, Make MoreOpenHands (formerly OpenDevin) is an AI-powered platform designed for software development, enabling agents to perform tasks that human developers usually handle, like modifying code, running commands, browsing the web, and even using code snippets from StackOverflow.audiocraft: library for audio processing and generation with deep learningAudioCraft is a PyTorch-based library developed by Facebook for deep learning research in audio generation. It includes models like MusicGen for controllable text-to-music generation, AudioGen for text-to-sound generation, and EnCodec for high-fidelity audio compression.MidJourney-Styles-and-Keywords-ReferenceA reference containing Styles and Keywords that you can use with MidJourney AI. There are also pages showing resolution comparison, image weights, and much more.jepa: PyTorch code and models for V-JEPA self-supervised learning from videoInstead of relying on labeled data, it predicts features from video frames, learning in a completely unsupervised manner. It processes video content to capture spatio-temporal patterns and trains a lightweight model to handle various downstream video and image tasks without adapting the core model.chat-with-mlx: An all-in-one LLMs Chat UI for Apple Silicon Mac using MLX Framework"Chat with MLX" is an all-in-one chat playground designed for Apple Silicon Macs, utilizing the Apple MLX framework. It allows users to securely chat with various AI models and integrate open-source models from platforms like HuggingFace.📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{line-height:0;font-size:75%} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

Transform your database into your AI platformAI_Distilled #74: Unlocking 7B+ language models in your browser: A deep dive with Google AI Edge's MediaPipe200+ hours of research on AI tools & hacks packed in 3 hoursThis free 3-hour Training on AI & ChatGPT (worth $399) will help you become a master of 20+ AI tools & prompting techniques and save 16 hours/week.Get it now for absolutely free! (for first 100 users only) 🎁You will learn how to:- Build business that make $10,000 by just using AI tools- Make quick & smarter decisions using AI-led data insights- Write emails, content & more in seconds using AI- Solve complex problems, research 10x faster & save 16 hours every weekRegister & save your seat now! (100 free seats only)SponsoredWelcome to AI_Distilled. Today, we’ll talk about:Awesome AI:LM Studio - Discover, download, and run local LLMsPainless Data Extraction and Web AutomationFleak AI Serverless API BuilderListen to Actual Clients' FeedbackTheysaid - Conversational AI SurveysMasterclass:Unlocking 7B+ language models in your browser: A deep dive with Google AI Edge's MediaPipeDeploying Attention-Based Vision Transformers to Apple Neural EngineMistral-NeMo: 4.1x Smaller with Quantized MinitronConnect the Amazon Q Business generative AI coding companion to your GitHub repositoriesAugmenting recommendation systems with LLMsHackHub:high-performance, multiplayer code editor from the creators of Atom and Tree-sitter.Multi-Platform Package Manager for Stable DiffusionSharpen your low-resolution pictures with the power of AI upscalingTransform your database into your AI platformLarge language model series developed by Qwen team, Alibaba Cloud.Cheers!Shreyans SinghEditor-in-Chief, Packt💻 Awesome AI: Tools for WorkLM Studio - Discover, download, and run local LLMsLM Studio 0.3.0 is a major update to the local LLM desktop application that enhances its offline capabilities with new features. Users can now chat with documents, using either full document context or "Retrieval Augmented Generation" (RAG) for longer texts. The update also introduces an OpenAI-like JSON output API, customizable UI themes, and automatic hardware detection for optimal performance.Painless Data Extraction and Web Automation (agentql.com)AgentQL is a powerful tool for data extraction and web automation that uses AI to reliably find and interact with web elements, even as websites change. Unlike traditional methods that rely on fragile XPath or DOM selectors, AgentQL allows users to locate elements using natural language descriptions, making it easier to automate tasks like filling forms, gathering data, and conducting end-to-end testing.Fleak AI Workflows. Simplified | Serverless API Builder | fleak.aiFleak is a low-code, serverless API builder designed for data teams to quickly and easily create, integrate, and scale AI and data workflows without managing any infrastructure. It allows users to configure and deploy workflows in minutes, seamlessly integrating with tools like large language models, vector databases, and modern storage technologies.Listen to Actual Clients' Feedback | Seven24 AISeven24 helps you capture and act on user feedback with ease. Integrate their tool into your product to collect feedback via text or voice, and their AI transforms this feedback into actionable tasks. With features like sentiment analysis, you can boost positive reviews and address issues quickly.Theysaid - Conversational AI SurveysTheySaid offers the world’s first conversational AI survey, designed to significantly increase response rates and improve customer engagement. By integrating seamlessly with your existing tech stack, the AI tool generates personalized survey questions based on your website content and follows up with users through conversational interactions.🔛 Masterclass: AI/LLM TutorialsUnlocking 7B+ language models in your browser: A deep dive with Google AI Edge's MediaPipeGoogle AI Edge's MediaPipe has developed a new system that allows large language models (LLMs) to run directly in web browsers, overcoming memory and performance limitations. By using WebAssembly and WebGPU, MediaPipe can now load and execute models like Gemma 1.1 with 7 billion parameters, which was previously unfeasible in-browser. The approach includes breaking down models into manageable parts and leveraging efficient memory usage techniques to handle the massive size of LLMs.Deploying Attention-Based Vision Transformers to Apple Neural EngineThe concept of Vision Transformers (ViTs) was introduced to leverage transformer models, which were originally used in natural language processing, for image recognition tasks. Unlike traditional Convolutional Neural Networks (CNNs), Vision Transformers process images by dividing them into smaller patches and applying attention mechanisms. This approach can handle various computer vision tasks such as image classification and object detection more effectively.Mistral-NeMo: 4.1x Smaller with Quantized MinitronNVIDIA's Minitron technique makes large language models (LLMs) like Mistral-NeMo smaller and more efficient by removing less critical parts and retraining them. This process reduces the models' sizes while keeping their performance high. The Minitron version of Mistral-NeMo, for instance, shrinks the model from 12 billion to 8 billion parameters. Combining Minitron with 4-bit quantization further compresses these models, allowing them to run on smaller GPUs and reducing operational costs.Connect the Amazon Q Business generative AI coding companion to your GitHub repositoriesYou can link Amazon Q Business, an AI-powered assistant, to your GitHub repositories using the Amazon Q GitHub (Cloud) connector. This setup allows you to use natural language queries to access information like commits, issues, and pull requests from your GitHub repositories. By integrating this tool, your development team can boost productivity, reduce context switching, and quickly retrieve information from your GitHub data through a conversational interface.Augmenting recommendation systems with LLMsLarge language models (LLMs), like Google's PaLM, can significantly enhance recommendation systems by integrating advanced AI capabilities. By incorporating LLMs into the recommendation pipeline, you can improve features like conversational recommendations, sequential recommendations based on user activity, and rating predictions. LLMs can interactively suggest items, understand the sequence of user preferences, and predict ratings with high accuracy.🚀 HackHub: AI Toolszed-industries/zedZed is a high-performance, multiplayer code editor developed by the team behind Atom and Tree-sitter. It can be installed on macOS and Linux directly or through package managers, though it’s not yet available for Windows or web platforms.LykosAI/StabilityMatrixStability Matrix is a multi-platform tool designed for managing Stable Diffusion Web UI packages across Windows, Linux, and macOS. It features a customizable interface with a syntax-highlighted terminal, a model browser for importing models from CivitAI and HuggingFace, and a shared model directory for all packages.Lucchetto/SuperImageSuperImage is an Android app that uses AI to enhance low-resolution images by upscaling them to higher resolutions. Built with the MNN framework and Real-ESRGAN, it processes images in tiles on the device's GPU, merging them into a high-resolution final image. It requires Android 7 or above and support for Vulkan or OpenCL.superduper-io/superduperIntegrate AI models and machine learning workflows with your database to implement custom AI applications, without moving your data. Including streaming inference, scalable model hosting, training and vector search.QwenLM/Qwen2Qwen2 is a suite of advanced language models available in various sizes, including up to 72 billion parameters. It offers state-of-the-art performance in tasks like coding and math, and supports up to 128K tokens for extended context. The models are pretrained and instruction-tuned, and they are available for use through Hugging Face and ModelScope.📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

Master the art of engineering Large Language Models from concept to productionAI_Distilled: Special IssueLLM Engineer's Handbook: Master the art of engineering LLMs from concept to productionCHECK IT OUTWelcome to a special edition of AI Distilled!In an era where AI is reshaping industries and redefining possibilities, staying ahead of the curve isn't just an advantage—it's a necessity.Whether you're a seasoned data scientist, a cybersecurity expert, or a curious developer looking to harness the power of Large Language Models (LLMs), this curated collection is designed to empower you with the latest insights and practical knowledge.📚 Inside This Special Issue:Master the art of prompt engineering and unlock AI's creative potentialDive deep into NLP, from foundational concepts to cutting-edge LLMsLeverage ChatGPT for enhanced cybersecurity measuresBuild powerful, data-driven applications using LlamaIndex and RAG techniquesGain insights from Supreet Kaur's expertise on choosing and implementing open-source LLMs🎙️ Don't Miss Out: Join Supreet Kaur's Free AMA Session!Whether you're looking to enhance your AI skills, stay ahead in your field, or explore new horizons in technology, this collection has something for everyone. Let's embark on this AI journey together and shape the future of technology!Happy learning,Shreyans SinghEditor in ChiefExpert Insight: Supreet Kaur"Navigating the LLM Landscape: Key Insights from Supreet Kaur's '100 Days of LLMs'"Supreet Kaur, a LinkedIn Top Voice 2024 and Data & AI Solutions Architect, has been sharing valuable insights on Large Language Models (LLMs) in her "100 Days of LLMs" series. Here are the key takeaways for AI professionals:Selecting the Appropriate ModelWhen deciding between small and large language models, Kaur emphasizes considering:📌Computational resources📌Use case complexity📌Real-time processing needsFor targeted applications with cost constraints, she highlights Microsoft's Phi-3 as a notable small model option.Leveraging Retrieval Augmented Generation (RAG)Kaur introduces RAG as a game-changing technique that combines generative AI with real-time information retrieval. This approach is particularly valuable in industries like fintech, where up-to-date information is crucial for decision-making.Rethinking Evaluation MetricsDrawing from her experience in text labeling automation, Kaur advocates for looking beyond conventional metrics. She suggests incorporating feedback from subject matter experts who will be using the model in practice, providing a more holistic evaluation.The Potential of AI AgentsKaur describes AI agents as autonomous software entities that can perform tasks on behalf of users or other programs. These "virtual interns" represent a promising frontier for enhancing productivity and tackling complex challenges across various domains.Effective LLM Evaluation StrategiesKaur outlines three key approaches for evaluating LLMs:📌Performance Metrics: Focusing on relevance, coherence, and groundedness📌Benchmark Testing: Comparing model versions under consistent conditions📌User Feedback: Gathering insights on real-world performanceShe also notes that platforms like Microsoft Azure offer tools to streamline this evaluation process.In conclusion, Kaur's advice helps people use AI language models better in real-world situations. She focuses on practical tips and new ideas that can help businesses make the most of this exciting technology.Join Supreet Kaur, LinkedIn Top Voice 2024 and AI Solutions Architect, for an insightful AMA session focused on leveraging open-source Large Language Models (LLMs) in real-world AI projects.FREE RegistrationUnlocking the Secrets of Prompt EngineeringLearn how to use AI writing tools for various tasks, from creating content to developing chatbots.The book covers:1. Basics of prompt engineering2. How to write effective prompts for AI3. Using AI for different types of writing4. Advanced uses like podcast creation and chatbot developmentGet eBook For $35.99 $24.99Mastering NLP from Foundations to LLMsLearn how to work with NLP using Python, focusing on both traditional techniques and modern LLMs like GPT.It covers the mathematical basics such as linear algebra and probability, and then moves on to more advanced topics like text classification, preprocessing, and deep learning models.You will find detailed Python code examples to help you build and implement ML models.Get eBook For $42.99 $29.99ChatGPT for Cybersecurity CookbookThis is a practical guide for leveraging AI, particularly ChatGPT, in cybersecurity.It provides step-by-step recipes to automate tasks like penetration testing, vulnerability assessments, and threat detection using the OpenAI API and Python programming.The book is designed for both beginners and professionals, offering tools to streamline cybersecurity workflows and improve efficiency through AI.Get eBook For $39.99 $27.98Building Data-Driven Applications with LlamaIndexLearn how to enhance their LLM applications using RAG.It teaches you how to overcome common limitations in LLMs, like memory constraints, prompt size, and inaccurate responses.You'll learn to build, customize, and deploy LlamaIndex projects, which allow better data ingestion, indexing, and querying.Get eBook For $35.99 $24.99More Titles for You$21.99 $31.99$24.99 $35.99$15.99 $23.99📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{line-height:0;font-size:75%} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

Nvidia GTC, OpenAI’s latest breakthrough, and what’s next in GenAI!AI_Distilled #87: What’s New in AI This WeekAI threats are evolving—here’s how to build unbreakable cyber resilience and fight misinformation before it spreads.READ FULL ARTICLEThe AI world isn’t slowing down, and neither are we. We’re back with another issue to keep you in the loop. High-stakes corporate moves and creative AI applications getting Hollywood’s attention: there’s plenty to catch up on this week. And it’s all here, concisely curated for you. Let’s get to it!LLM Expert Insights Team,PacktIn today's issue:Recent Developments – Baidu’s AI advancements, Intel’s strategy shift, Google-MediaTek partnership, US blocks DeepSeek, AI-driven cyber defense, SoftBank’s $6.5B Ampere dealNvidia GTC 2025 – Blackwell Ultra AI chip, Dynamo inference software, Cisco-Nvidia Secure AI FactoryHollywood & AI – Russo Brothers explore AI in Marvel, copyright law debatesGame-Changing AI Tools – OpenAI’s ChatGPT Connectors, Nvidia & MIT’s new image generation tech📰RECENT DEVELOPMENTSBold acquisitions and groundbreaking AI advancements: the tech world is buzzing with big moves and new strategies. Here are a few intriguing shifts turning heads this week:Baidu steps up the AI game with new models and free chatbot accessBaidu has launched a new AI reasoning model, X1, and its latest foundation model, Ernie 4.5, while making its Ernie Bot chatbot free for individual users ahead of schedule. The X1 model is said to offer performance comparable to DeepSeek’s efficient model at a lower cost, while Ernie 4.5 reportedly outperforms OpenAI’s GPT-4.5 across various benchmarks. The move comes as Chinese tech companies rush to enhance their AI platforms following DeepSeek’s groundbreaking open-source release.Intel's new CEO targets AI and chip manufacturing revampIntel’s incoming CEO, Lip-Bu Tan, is preparing to restructure the company’s AI and chip manufacturing strategies, aiming to enhance efficiency and reclaim Intel’s standing in the semiconductor industry. His plans include streamlining middle management, improving Intel’s Foundry operations (which makes chips for other design companies such as Microsoft and Amazon), and developing AI chips using advanced 18A process technology. Tan also aims to attract major customers like Nvidia and Broadcom, positioning Intel for a stronger future in AI-driven chip manufacturing.Google partners with MediaTek for next-gen AI chipsGoogle is reportedly collaborating with MediaTek to develop the next generation of TPUs, expected to be produced next year. The partnership is driven by MediaTek’s strong ties with TSMC and its lower production costs compared to Broadcom, which Google also partners with for AI chip development. Google’s TPU chips play a critical role in its AI strategy, powering services like Google Search, YouTube, and Gemini AI models.US Commerce Department blocks DeepSeek over data privacy concernsThe US Commerce Department has prohibited the use of the Chinese AI model DeepSeek on government devices, citing concerns over data privacy and potential exposure of sensitive information. The ban, communicated through mass emails to staff, aligns with broader legislative efforts by Congress members pushing to restrict DeepSeek’s access on government-issued equipment due to fears of data exploitation by the Chinese government.Sophos leverages multimodal AI for advanced cyber defenseAt the 2024 Virus Bulletin conference, Sophos Principal Data Scientist Younghoo Lee presented research on using multimodal AI to enhance spam, phishing, and web content detection. Unlike traditional models, multimodal AI analyzes both text and visuals simultaneously, identifying sophisticated threats by understanding how legitimate and malicious content differ across multiple data types. Its capabilities include detecting phishing tactics through text analysis, brand verification, and advanced URL screening.Prompt Security unveils AI safeguards to prevent unauthorized data accessPrompt Security has introduced new authorization features to enhance security and control over generative AI applications within enterprises. The system provides real-time prevention of unauthorized data access by analyzing user identity and request context, ensuring AI tools like Copilot and Google Gemini adhere to existing security policies. Integrated with identity providers like Okta and Microsoft Entra, the platform offers granular policy enforcement, flexible redaction options, and comprehensive audit logging to protect sensitive corporate data.SoftBank to acquire Ampere Computing in $6.5 billion AI-focused dealSoftBank Group has announced its acquisition of Ampere Computing, a startup known for its Arm-based server chips, for $6.5 billion. The deal, expected to close in the second half of 2025, will see Ampere operating as an independent subsidiary with its headquarters in Santa Clara, California. SoftBank aims to enhance its AI infrastructure investments, building on partnerships like its recent collaboration with OpenAI and participation in the $500 billion Stargate AI project.⚡NVIDIA GTC 2025Nvidia’s GTC 2025 conference is making waves with several major announcements aimed at revolutionizing AI infrastructure, performance, and security. Take a look at our roundup of the most significant updates coming out of the event.Nvidia launches Blackwell Ultra AI chip to revolutionize AI processingAt GTC 2025, Nvidia unveiled its Blackwell Ultra AI chip, which offers 1.5 times the performance of its predecessor and significantly boosts AI processing capabilities. The chip powers Nvidia’s new GB300 superchip, designed for AI systems used by major companies like Amazon, Google, Microsoft, and Meta. Nvidia claims the Blackwell Ultra, paired with its DGX SuperPod AI supercomputer, dramatically enhances AI reasoning capabilities, delivering faster and more efficient responses than previous models.Nvidia Dynamo: New open-source AI inference software for enhanced efficiencyNvidia has launched Dynamo, an open-source AI inference software designed to improve the efficiency and scalability of AI reasoning models within AI factories. By using techniques like disaggregated serving, which separates processing and generation tasks across GPUs, Dynamo promises to double AI performance and revenue generation while minimizing operational costs. The software is compatible with popular frameworks like PyTorch and NVIDIA TensorRT-LLM, making it accessible to enterprises, cloud providers, and AI innovators worldwide.Cisco and Nvidia partner to launch Secure AI Factory for enterprise AI infrastructureCisco and Nvidia have introduced the Cisco Secure AI Factory, a comprehensive AI architecture package designed to enhance AI networking security and efficiency. The solution integrates Cisco’s Hypershield and AI Defense packages with Nvidia DPUs, SuperNICs, and enterprise storage from partners like Pure Storage and NetApp. Aimed at safeguarding AI development, deployment, and operations, the platform offers flexible deployment models and reference architectures for industries including finance, healthcare, and manufacturing.🤖 GAME-CHANGING AI TOOLSFrom boosting enterprise productivity to making AI more accessible to everyone, take a look at the most compelling tools and innovations sparking conversations right now.OpenAI to pilot ChatGPT Connectors for Google Drive and Slack dOpenAI is set to launch a beta feature called ChatGPT Connectors, allowing business users to link Google Drive and Slack accounts to ChatGPT. This integration aims to enhance the chatbot’s ability to answer queries using internal files, presentations, spreadsheets, and Slack conversations. OpenAI plans to expand this feature to other platforms like Microsoft SharePoint and Box.Nvidia and MIT unveil ‘HART’ for faster, high-resolution image generationNvidia and MIT have introduced a new tool called ‘HART’ that merges the strengths of diffusion models and autoregressive models into a unified approach. Designed to generate highly realistic images more efficiently than some current models, HART delivers high-resolution results with minimal steps. Its scalability is projected to be exponential, with future integration plans for video generation and audio prediction tasks.Oracle’s AI Agent Studio empowers Fusion Cloud users with custom AI agentsOracle’s AI Agent Studio is now enhancing the Oracle Fusion Cloud Applications Suite by allowing businesses to create and manage AI agents tailored to their needs. With tools for agent orchestration, LLM integration, and data validation, the platform promises streamlined workflows while ensuring security and reliability. However, areas like governance and privacy compliance still require further attention.SAP introduces 'Joule for Developer' to enhance AI-driven developmentSAP has launched 'Joule for Developer,' a new AI co-pilot aimed at improving SAP Build tools for developers by integrating purpose-built LLMs. The tool offers intelligent suggestions, automates tasks like documentation and sample data generation, and supports code optimization and process automation. With seamless integration across SAP Build tools and SAP Business Application Studio, SAP aims to empower developers to build more efficient, innovative, and secure applications. Looking ahead, SAP plans to enhance this tool with AI agents offering improved data security and AI-compliant platforms.Get lifetime access to top AI tools with 1min.AI1min.AI is an AI platform offering lifetime access to popular tools like GPT-4, Gemini, and other leading AI solutions for a one-time payment of $79.97. Equipped with cutting-edge tools for editing and various content-related tasks, 1min.AI promises a powerful boost to your AI toolkit, helping you stay updated with the latest trends in AI technology. The deadline to secure a lifetime subscription to 1min.AI for just $79.97 is March 30 at 11:59 p.m. PT.🎬 HOLLYWOOD AND AIHollywood’s relationship with AI continues to evolve, balancing both excitement and concern about the ethical implications. How are industry leaders navigating the intersection of technology and artistic expression?Russo Brothers explore AI’s role in future Marvel projectsThe Russo Brothers, known for their groundbreaking visuals in Marvel films, have shared their thoughts on the potential use of AI in Avengers: Doomsday and Avengers: Secret Wars. They believe AI could enhance the creative process by leveraging advanced editing technology to deliver a superior cinematic experience. However, the challenge lies in responsibly integrating AI into their filmmaking approach.Hollywood opposes proposed AI copyright law changesLeading Hollywood figures have addressed an open letter to the administration, raising concerns about proposed changes to copyright laws affecting AI training. They argue that relaxing these laws could harm the creative industry, which employs thousands of Americans, by compromising the integrity of original content and artistic expression.Learn enterprise patterns, key design principles, and proven architectures for building AI agents with LangChain and LangGraph.Preorder Generative AI with LangChain today!PRE-ORDER NOW!📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0}#converted-body .list_block ol,#converted-body .list_block ul,.body [class~=x_list_block] ol,.body [class~=x_list_block] ul,u+.body .list_block ol,u+.body .list_block ul{padding-left:20px} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

AI Code Review for Developers | TragAI_Distilled #78: Customize how Claude responds: Concise, Explanatory, or FormalLearn the Roadmap to making $100k using LinkedIn & AI (for free)🚀In just 90 minutes, you’ll learn how to:👉 Automate lead generation to grow your business effortlessly.👉 Master LinkedIn's $100K strategy to increase revenue while saving time.👉 Use AI to secure high-paying roles, bypassing endless applications.Join Vaibhav Sisinty, a LinkedIn influencer with over 400K followers, who’s transformed the LinkedIn strategies of over 200,000 professionals. Normally valued at $399, this workshop is free for the first 100 readers.Claim Your Free Spot Now (Only 100 seats available!)Welcome to AI_Distilled. Today, we’ll talk about:TechwaveCustomize how Claude responds: Concise, Explanatory, or FormalRunwayML: Introducing FramesAnthropic introduces the Model Context Protocol: SmolVLM - small yet mighty Vision Language ModelCursor announces new code editor UI and agentAwesome AI:Paperguide: AI Research Assistant & Chat with PDFCapGo AI: Spreadsheet That Fills ItselfAI Code Review for Developers | TragConversational AI Survey with Real-time Follow upsSagaLabs: Earn 200x More with In-context AI translation from the worldMasterclass:ControlNets for Stable Diffusion 3.5 Large — Stability AIAutomatically generating cloud configurations: Introducing RAGformationBoost your Continuous Delivery pipeline with Generative AI | Google CloudCreating with Video to Video on Gen-3 Alpha and Turbo – RunwayModel-Based Transfer Learning for Contextual Reinforcement LearningHackHub:Andrew Ng releases an open-source Python framework to swap between LLMs with one line of codeOpenInterpreter/open-interpreter: A natural language interface for computersItzCrazyKns/Perplexica: Perplexica is an AI-powered search engine. It is an Open source alternative to Perplexity AIsouzatharsis/podcastfy: An Open Source Python alternative to NotebookLM's podcast feature: Transforming Multimodal Content into Captivating Multilingual Audio Conversations with GenAIblack-forest-labs/flux: Official inference repo for FLUX.1 modelsCheers!Shreyans SinghEditor-in-Chief, PacktScale your scrapers with Apify’s Black Friday Boost planGet a 30% prepaid usage bonus on Apify this Black Friday. Scrape data for app integrations, performance tracking, competitive research, or custom pipelines. Use pre-built scrapers, build your own from scratch, or use quick-start code templates. The Boost plan ends December 5 - grab it while you can!Claim your bonus now⚡ TechWave: AI/GPT News & AnalysisCustomize how Claude responds: Concise, Explanatory, or FormalAnthropic has introduced a new feature for its Claude AI assistant that allows users to customize its writing style to match their own or adjust it for specific tasks. Users can choose from three preset styles—Formal, Concise, and Explanatory—or create personalized styles by uploading sample text for Claude to mimic. This feature aims to make interactions feel more natural and tailored, whether for technical documents, professional emails, or casual chats.RunwayML: Introducing FramesRunway's new image generation model, Frames, offers advanced stylistic control and visual fidelity, allowing creators to design consistent yet creatively flexible visuals. Integrated into Gen-3 Alpha and the Runway API, Frames helps users craft detailed aesthetic worlds, from cinematic portraits to retro-inspired designs. Frames aims to redefine creative workflows by enabling precise and imaginative visual storytelling.Anthropic introduces the Model Context Protocol: Anthropic has introduced the Model Context Protocol (MCP), an open-source standard aimed at improving how AI assistants access and use data from various sources, like business tools and content repositories. MCP enables two-way connections between AI models and data systems through "MCP servers" and "MCP clients," simplifying integration and reducing the need for custom connectors. promising to create more seamless and scalable AI integrations, MCP faces competition from proprietary alternatives like OpenAI’s "Work with Apps,".SmolVLM - small yet mighty Vision Language ModelSmolVLM is a highly efficient and compact 2-billion-parameter Vision-Language Model (VLM) that delivers state-of-the-art performance for its size and memory usage. Designed for speed, memory efficiency, and ease of customization, SmolVLM is fully open-source under the Apache 2.0 license, with tools, training recipes, and datasets readily available. Its three variants—Base, Synthetic, and Instruct—support fine-tuning and out-of-the-box applications. By optimizing image token encoding and leveraging innovative architecture, SmolVLM runs effectively on smaller devices like laptops, offering fast inference and low GPU memory usage.Cursor announces new code editor UI and agentCursor's 0.43 update transforms the AI-powered code editor into a more efficient and developer-friendly tool. Key features include a unified workspace with the redesigned Composer UI, advanced automation for debugging and package installation via the Composer Agent, and enhanced semantic search for faster, context-aware results. The update also introduces proactive debugging with the experimental BugFinder tool, visual cues for easier file management, and context-aware coding suggestions.💻 Awesome AI: Tools for WorkPaperguide: AI Research Assistant & Chat with PDFCapGo AI: Spreadsheet That Fills ItselfAI Code Review for Developers | TragConversational AI Survey with Real-time Follow upsSagaLabs: Earn 200x More with In-context AI translation from the world🔛 Masterclass: AI/LLM TutorialsControlNets for Stable Diffusion 3.5 Large — Stability AIStable Diffusion 3.5 Large introduces three new ControlNets—Blur, Canny, and Depth—designed to enhance image generation precision. Blur enables high-fidelity upscaling for detailed visuals, Canny uses edge maps for structured illustrations, and Depth leverages depth maps for architectural and 3D applications. These models are free for non-commercial and small-scale commercial use.Automatically generating cloud configurations: Introducing RAGformationRAGformation is an open-source AI tool designed to simplify cloud configuration by automating the selection of services, cost estimation, and architecture design. Using natural language input, it generates tailored cloud setups, including visual flow diagrams, pricing details, and a comprehensive blueprint. Powered by Retrieval-Augmented Generation (RAG) and tools like LlamaIndex and Pinecone, RAGformation dynamically adjusts recommendations based on user preferences and budgets.Boost your Continuous Delivery pipeline with Generative AI | Google CloudGenerative AI, such as Google Cloud's Gemini models, enhances software development by automating repetitive tasks and improving code quality throughout the development lifecycle. Beyond assisting in coding within IDEs, AI can streamline continuous delivery pipelines by automating code reviews, generating release notes, and detecting potential issues early. For example, integrating Gemini into a CI/CD pipeline allows developers to receive AI-driven feedback on pull requests and summaries of code changes, reducing manual effort and boosting productivity. Tools like the "friendly-cicd-helper" demonstrate how AI can complement traditional processes, freeing developers to focus on strategic tasks while maintaining high-quality standards.Creating with Video to Video on Gen-3 Alpha and Turbo – RunwayThe Gen-3 Alpha and Turbo models offer an enhanced "Video to Video" feature, allowing users to transform the style of videos using text prompts. The Turbo model is faster and more cost-effective, supporting resolutions up to 1280x768 and videos of up to 20 seconds. To use this feature, select a model, upload a supported video, and draft a detailed prompt to define the desired style. Additional settings, like structure transformation and aspect ratio, allow for customization. Once configured, the tool generates stylized videos, with results saved in the Generative Video folder for easy access.Model-Based Transfer Learning for Contextual Reinforcement LearningThis paper introduces Model-Based Transfer Learning (MBTL), a framework to improve generalization in contextual reinforcement learning (RL). Traditional RL approaches often fail with minor environmental changes, and existing training methods are either too resource-intensive or prone to negative transfer. MBTL addresses this by modeling generalization performance with Gaussian processes and linear functions to predict and minimize performance gaps when transferring policies to new tasks. By integrating these models with Bayesian optimization, MBTL strategically selects training tasks, achieving up to 50x better sample efficiency in benchmarks like urban traffic. This approach paves the way for more reliable and efficient RL training methods.🚀 HackHub: AI ToolsAndrew Ng releases an open-source Python framework to swap between LLMs with one line of codeOpenInterpreter/open-interpreter: A natural language interface for computersItzCrazyKns/Perplexica: Perplexica is an AI-powered search engine. It is an Open source alternative to Perplexity AIsouzatharsis/podcastfy: An Open Source Python alternative to NotebookLM's podcast feature: Transforming Multimodal Content into Captivating Multilingual Audio Conversations with GenAIblack-forest-labs/flux: Official inference repo for FLUX.1 models📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

Google announces Veo and Imagen 3: new video and image generation modelsAI_Distilled #79: Sam Altman announces "12 days of OpenAI"Learn Million Dollar AI Strategies & Tools in this 3 hour AI Training for Free.If you are not an AI-powered professional today, you will either:-Get replaced by a person who uses AI-Face a slow career growth & lower salary-Keep spending 10s of hours on tasks that can be done in 10 minutes.Best thing? We’re running the Black Friday Sale so you can get it for absolutely free (for the first 100 readers).Save your seat now (Offer valid for 24 hours only)Welcome to AI_Distilled. Today, we’ll talk about:TechwaveSam Altman announces "12 days of OpenAI"Google announces Veo and Imagen 3: new video and image generation modelsDeepMind Genie 2: generate interactive worlds that look like video gamesIntel data scientist's survival guide to GenAINvidia launches Ingest: Multimodal PDF Data ExtractionAwesome AI:Polymet - Idea to prototype within secondsClipAnything - Choppityfal.aiEarkick - Your Personal AI ChatbotOuterbase | The interface for your databaseMasterclass:Voice Trigger System for SiriAlign Meta Llama 3 to human preferences with DPOAn Intuitive Intro to RLEnhancing LLMs with Structured Outputs and Function CallingSafely repairing broken builds with MLHackHub:Agents for software developmentOpen-source LLM app development platformbuild, manage & run useful autonomous agentsUnderstand Human Behavior to Align True NeedsGenerative models for conditional audio generationCheers!Shreyans SinghEditor-in-Chief, Packt⚡ TechWave: AI/GPT News & AnalysisSam Altman announces "12 days of OpenAI"OpenAI is celebrating with a special event called "12 Days of OpenAI," where, for twelve days, the company will reveal new models, features, and updates via livestreams. Anticipated reveals include full release of its o1 reasoning model, updates on its voice modes, including a festive Santa voice, a new AI agent called Operator, a web browser, a desktop app update, and advancements in AI-generated music and vision fine-tuning. Notably, OpenAI may also introduce new AI chips and even GPT-5, which promises improved reasoning and customization.Google announces Veo and Imagen 3: new video and image generation modelsGoogle Cloud has introduced two advanced generative AI models, Veo and Imagen 3, on its Vertex AI platform. Veo allows businesses to generate high-quality videos from simple text or image prompts, transforming creative assets into dynamic visuals quickly and affordably. Imagen 3, launching next week, creates highly realistic images from text prompts, offering more detail and fewer visual artifacts than previous models. Both models are built with safety features, such as digital watermarking and safety filters, to ensure responsible use.DeepMind Genie 2: generate interactive worlds that look like video gamesDeepMind has introduced Genie 2, an advanced AI model capable of generating interactive 3D worlds that resemble video games. Unlike previous models, Genie 2 can create dynamic environments from just a single image and a text description, allowing users to interact with the scene, like jumping or swimming. The model simulates object interactions, physics, and animations, and can remember parts of the world even when they’re not visible, offering a more consistent and realistic experience. While not designed for full gaming experiences, Genie 2 is a tool for research, creative prototyping, and evaluating AI agents.Intel data scientist's survival guide to GenAIWhile GenAI tools can produce impressive results, they heavily rely on clean, well-structured data and insightful interpretation—areas where data scientists excel. Your expertise in data analysis, modeling, and statistical methods ensures that these models can make accurate, actionable predictions. GenAI platforms need data scientists to optimize and evaluate models, enhance their performance, and ensure their deployment is successful. Tools like Modin, Intel-optimized frameworks, and MLflow help streamline the process, making data preparation, model training, and deployment more efficient, particularly when working on Intel hardware.Nvidia launches Ingest: Multimodal PDF Data ExtractionNVIDIA-Ingest is a powerful microservice for extracting and processing content from documents like PDFs, Word, and PowerPoint files. It can analyze and separate text, images, tables, and charts, delivering them in a structured JSON format. Using NVIDIA's advanced tools, including OCR and AI-driven parsing, it enables efficient data processing for downstream applications like generative AI or embedding storage in vector databases like Milvus. It supports flexible workflows and can handle tasks like splitting documents, generating embeddings, and transforming data💻 Awesome AI: Tools for WorkPolymet - Idea to prototype within secondsPolymet is an AI-powered tool that helps users quickly turn ideas into prototypes by generating designs and production-ready code in seconds. Users can describe what they need, iterate on the design with their team, and then export the code and designs, which can easily integrate with tools like Figma and existing codebases.ClipAnything - ChoppityChoppity is an AI-powered video editing tool that allows users to quickly find and clip moments from any video using visual, audio, and sentiment analysis. With its "ClipAnything" feature, users can search for specific parts of a video, such as key events, people, or emotions, without having to manually review hours of footage.fal.aiFal.ai is a generative media platform designed for developers to create and deploy AI-powered applications, particularly focused on text-to-image models. It offers fast, cost-effective inference with models like FLUX.1 and Stable Diffusion, optimized for various creative tasks.Earkick - Your Personal AI ChatbotEarkick is an AI-powered mental health app that helps users track and improve their emotional well-being in real time through a personal chatbot named Panda. Earkick tracks mental readiness, mood, and calmness, while providing daily insights, breathing techniques, and guided self-care sessions.Outerbase | The interface for your databaseOuterbase is an AI-powered platform that simplifies working with databases for engineers, researchers, and analysts. It supports SQL and NoSQL databases, allowing users to manage data securely while using AI tools to write queries, fix mistakes, and generate charts and visualizations instantly. Outerbase's table editor, dashboards, and data catalog help users organize, analyze, and share insights efficiently.🔛 Masterclass: AI/LLM TutorialsVoice Trigger System for SiriApple's voice trigger system for Siri includes a first-stage low-power detector to identify potential triggers, and a second-stage, high-precision model to confirm the trigger. It also incorporates speaker identification to ensure the device responds only to its primary user. This sophisticated setup addresses challenges like background noise and phonetically similar words while maintaining power efficiency and privacy.Align Meta Llama 3 to human preferences with DPODPO involves fine-tuning a large language model (LLM) based on feedback from human annotators who rate or rank the model's responses according to desired values, such as helpfulness and honesty. SageMaker Studio provides the computational environment to fine-tune the model using Jupyter notebooks with powerful GPU instances, while SageMaker Ground Truth simplifies the process of gathering human feedback by managing workflows for data annotation. Together, they allow you to align the Llama 3 model’s responses with specific organizational values efficiently.An Intuitive Intro to RLReinforcement learning (RL) is a type of machine learning where an agent learns by interacting with its environment, making decisions, and receiving feedback in the form of rewards or penalties. The goal is to maximize cumulative rewards over time. The agent starts with little to no knowledge and improves through trial and error, learning from past experiences. In RL, actions taken by the agent change the state of the environment, and based on the rewards received, the agent adjusts its future actions. A key concept in RL is balancing exploration (trying new things) and exploitation (using known strategies for rewards).Enhancing LLMs with Structured Outputs and Function CallingEnhancing LLMs with structured outputs and function calling improves their ability to provide accurate and useful responses. Structured outputs ensure consistency and clarity by organizing information in a logical format, reducing ambiguity. Function calling allows LLMs to perform specific tasks, such as retrieving real-time data or executing external functions, making them more interactive and versatile. Combined with techniques like Retrieval-Augmented Generation (RAG), which integrates relevant external information into the model’s responses, these enhancements lead to more reliable, accurate, and contextually rich conversations with LLMs.Safely repairing broken builds with MLGoogle's engineers have developed a machine learning model called DIDACT to automatically repair broken code builds by analyzing historical data of build errors and their fixes. This model suggests potential fixes to developers directly within their Integrated Development Environment (IDE). In a controlled experiment, the use of these machine learning-suggested fixes improved productivity by reducing active coding and feedback time, and increasing the number of completed code changes.🚀 HackHub: AI ToolsAll-Hands-AI/OpenHandsOpenHands is an AI-powered platform designed to assist with software development, allowing agents to perform tasks similar to human developers. These agents can modify code, run commands, browse the web, call APIs, and even use resources like StackOverflow. OpenHands is easy to set up using Docker and can be run in various modes, including scriptable or interactive CLI.langgenius/difyDify is an open-source platform for developing AI applications, offering an intuitive interface that integrates workflows, agent capabilities, model management, and observability features. Dify's core features include a visual AI workflow builder, integration with numerous LLMs, agent tools, and a retrieval-augmented generation (RAG) pipeline for document handling.TransformerOptimus/SuperAGISuperAGI is an open-source framework designed for developers to create, manage, and run autonomous AI agents. It allows seamless operation of multiple agents simultaneously and provides tools to extend their capabilities. With features like graphical interfaces, performance telemetry, and integration with multiple vector databases, SuperAGI enables AI agents to efficiently handle tasks, learn from experience, and optimize token usage.lllyasviel/Paints-UNDOPaints-Undo is an open-source project that provides AI models designed to simulate the drawing process in digital art. By inputting a completed image, users can generate a sequence of steps showing how that image might have been created, mimicking the "undo" function in digital painting software.Stability-AI/stable-audio-toolsStable-Audio-Tools is an open-source library for working with audio generation models. It provides tools for training and running models that generate audio, including a Gradio interface for testing. Users can install the library via PyPI, and the repository includes scripts for both training models and performing inference.📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

DeepSeek open sources 5 repos for AGI, Helix and Engine AI’s humanoids gain more power, Agents in acAI_Distilled #84: Your AI News Fix!You can now train your own Reasoning model like DeepSeek-R1 locally with just 5GB VRAM. Unsloth is fully open-source and allows you to transform any open LLM like Llama 3.1 (8B) or Phi-4 (14B) into a reasoning model.GitHub repo: https://github.com/unslothai/unslothDeepSeek’s R1 research revealed an “aha moment” where R1-Zero autonomously learned to allocate more thinking time without human feedback by using Group Relative Policy Optimization (GRPO). Unsloth enhanced the entire GRPO process, making it use 90% less VRAM than all other implementations. This allows you to reproduce R1-Zero's "aha moment" on just 5GB of VRAM using Qwen2.5 (1.5B).Try Unsloth's free GRPO notebook with a free 16GB GPU: Llama 3.1 (8B) on ColabFor a Tutorial and GRPO notebooks featuring other models like Phi-4, visit Unsloth's docsIt looks like the AI giants are battling it out, with announcements on new models, Gen-AI capabilities for their flagship products, and research breakthroughs. But don’t you worry, we’ve got you. Here is your weekly digest!LLM Expert Insights Team,Packt📰 NewsDeepSeek open sources five repos for AGI in its OpenSourceWeekIn its OpenSource week, DeepSeek is making available five repos that form the building blocks of their online service. These repos include FlashMLA (efficient MLA decoding kernel for Hooper GPUs), DeepEP (EP communication library for MoE model training and inference), DeepGEMM (FP8 library supporting dense and MoE GEMMs), and DualPipe (a bidirectional parallelism algorithm), and Fire-Flyer File System (a parallel file system that utilizes the full bandwidth of modern SSDs and RDMA networks).Microsoft’s next generation of Phi-4 modelMicrosoft introduced Phi-4-multimodal and Phi-4-mini, the latest additions to Microsoft's Phi family of small language models (SLMs). Phi-4-multimodal handles speech, vision, and text concurrently, while Phi-4-mini is proficient in text-based tasks. Phi-4-multimodal is a 5.6B parameter model, and Phi-4-mini is a 3.8B parameter model. Both models are suitable for compute-constrained inference environments.Google announces public preview of Gemini Code AssistGoogle has made Gemini Code Assist available to individual developers for free public preview, with a liberal token window of 128K. This AI-coding assistant offers code completion, generation, and chat features in Visual Studio Code and JetBrains IDEs, similar to thosealready available in Firebase and Android Studio. And guess what, you have about 180,000 code completions every month! Insane! isn’t it? A similar tool, Gemini Code Assist for GitHub, is also available, providing AI-powered code reviews.Amazon introduces Gen-AI infused Alexa – Alexa+Amazon introduced Gen-AI-powered Alexa+ this week. It features agentic capabilities and is designed to be smarter than the original Alexa, with LLMs powering up its knowledge base. Designed to take actions, it can remember your specific needs and requirements, making your experiences more useful and personalized. Available on Echo devices, a new mobile app, and Alexa.com, it costs $19.99 per month but is free for Prime members.Claude’s 3.7 Sonnet Hybrid reasoning with extended thinking and Claude CodeAnthropic has announced Claude 3.7 Sonnet with hybrid reasoning capabilities. Users can now toggle between fast responses and extended thinking modes, with a budget of up to 128K tokens. Unlike other reasoning models, Claude is more focused on the real-world business applications of LLMs, rather than math and computer science competition tasks. Anthropic also introduced Claude Code, a command-line collaborative tool for agentic coding, currently available as a limited research preview.Alibaba’s open-sources thinking model QwQ-Max-PreviewAlibaba, through an announcement blog post created by QWQ-Max-Preview, unveiled the newest model in the Qwen series: QwQ-Max-Preview. It is built upon Qwen2.5-Max and excels in mathematics, coding, general tasks, and agentic workflows. The post also mentions future plans, which include the development of a dedicated app for Qwen Chat and smaller QwQ variants for local device deployment.Comet an agentic search browser by PerplexityPerplexity announced its agentic browser Comet in an X post. Built on the Chromium framework, Comet will integrate search and automate related tasks. It will also integrate deep research and real-time information processing. You can join the waitlist here.Perplexity also announced voice mode for its iOS app. Voice mode is expected to be shipped for Android and Mac apps in the coming days.Microsoft cancelling U.S. data center leases amid CEO Satya Nadella’s concerns about AGI milestonesA TD Cowen report states that Microsoft has pulled the plug on 200MW leases for at least two private data centers, withdrawn from around 500 leases, and reallocated a sizeable portion of its international spend to the US. In another development, CEO Satya Nadella, shared his thoughts on AGI hype. He opined that self-proclamation of AGI is useless and the true revolution, the real benchmark will be when we see growth in the GDP. “It can’t be just supply side,..,when the productivity goes up, and the economy is growing at a faster rate. When that happens… that’s to me is the moment,” he said.Alibaba to invest RMB 280 billion in AI and cloud computing infrastructureAlibaba plans to invest USD 53 billion over the next three years to scale up AI capabilities and cloud infrastructure, providing businesses with tools for innovation. CEO Eddie Wu sees AI as a "once-in-a-generation" opportunity. Cloud computing is Alibaba's main revenue driver in AI, with high demand for AI hosting services. Alibaba is integrating AI across its ecosystem to improve customer experiences, optimize business operations, and drive long-term growth.Apple makes $500 billion commitment to US’s future – Tim Cook, CEO, AppleApple plans to invest over $500 billion in the U.S. in the next four years, focusing on investments in AI, silicon engineering, manufacturing, and skills development. A new manufacturing facility will be opened in Houston for Apple Intelligence servers and the U.S. Advanced Manufacturing Fund will be doubled to $10 billion. A manufacturing academy will be established in Michigan, and R&D investments will expand across the U.S., creating about 20,000 jobs. Apple continues to support educational programs for hardware engineering and silicon chip design.SamA announces two new features for ChatGPT Plus and free usersOpenAI released research preview for GPT 4.5 this week to understand its strenght and limitations.In his X posts, OpenAI CEO, Sam Altman, announced DeepResearch for ChatGPT Plus users and Advanced Voice for GPT-4o mini.In another development, The Information reported that OpenAI plans to shift 75% of its data center capacity to StarGate, financed by SoftBank. This transition from Microsoft-owned data centers is expected to occur over the next five years.Meta for Education, a new mixed and virtual reality (VR/MR) offering, is now generally available. It provides educators with Meta Horizon-managed solutions, aimed at enhancing student engagement and knowledge retention through interactive VR/MR experiences.💻 Awesome AI: Tools for WorkAlibaba releases wan 2.1 family of video modelswan2.1 presents two versions of video generation models: a lightweight 1.3 billion parameter model suitable for laptops, and a robust 14 billion parameter model for higher performance. wan2.1 handles both text-to-video and image-to-video generation, providing resolution choices of 720p or 480p. It can simulate complex motion, capture intricate details, and generate multilingual text effects.Pika announces Pika 2.2, PikaFrames, andPikaswaps on XPikaswaps allows users to modify and replace objects in videos using video inpainting. It enables the swapping, erasing, and altering of objects while maintaining realistic visual consistency. Features include a brush tool, reference image uploads, and options to re-prompt or retry.Engine AI’s humanoid can perform complete front flipEngineAI has unveiled the world's first humanoid robot capable of performing a front flip. This achievement marks a significant advancement in humanoid robotics, showcasing improved agility and control. The robot's ability to execute complex acrobatic movements demonstrates advancements in AI-driven motion planning and real-time control systems.Grok3 voiceIn his X post, CEO, Elon Musk announced that xAI’s Grok3 has enabled conversation mode for Premium and SuperGrok users..Helix – A vision language action modelFigure AI’s Helix model is designed to bring humanoid robots into homes. It blends computer vision, language comprehension, and real-time motor control. Helix can adapt on the go, learn quickly with minimal training data, control multiple robots simultaneously, and handle thousands of household items. It runs on embedded low-power GPUs And can pick up virtually any small household object by voice command. 🛠️ HackhubMagma: A foundation model for multimodal AI agents across digital and physical worlds - Microsoft ResearchMicrosoft Research has introduced Magma, a foundation model for multimodal AI agents, to bridge the digital and physical worlds. Magma integrates diverse sensor data—such as vision, audio, and depth—enabling agents to perceive and interact with complex environments. It supports a wide range of tasks, from simple object recognition to intricate navigation and manipulation. It can create adaptable agents that can learn and generalize across various scenarios, enhancing robotics, AR/VR, and human–computer interaction.Meta’s ML GymMLGym is an open-source framework and benchmark designed to accelerate AI agent research. It aims to simplify the development, evaluation, and comparison of AI agents across diverse environments. By offering a standardized platform for researchers to conduct experiments, share results, and collaborate, MLGym will enable more efficient and reproducible research.PaliGemma 2 - New Instruction Vision Language Models by GooglePaliGemma2-Mix is a vision-language model based on the Gemma language model and SigLIP vision model. Optimized for efficiency and performance, the model is available on Hugging Face. It's designed for tasks requiring visual understanding and language generation, such as image captioning and visual question answering. The "mix" version provides a blend of pre-training and fine-tuning, offering a versatile and robust model.⚙️TechhubGibber link – AI Agent communication protocolGibber Link is an agent communication protocol that proposes the use of sound-level protocols instead of speech for efficient communication. This reduces compute costs by 90%, speeds up data transfer by 80%, and minimizes errors. The protocol automatically switches from speech to sound upon detecting another AI agent, enhancing clarity and enabling multimodal data exchange.Meta MotivoMeta Motivo is a tool by Meta Demolab that can be used for creating 3D character animations from audio inputs. It uses audio-driven motion generation and analyzes speech patterns to produce realistic facial expressions and body movements. Motivo employs a neural network trained on a large dataset of speech and motion capture data, enabling it to synthesize animations that synchronize with the audio.Introducing the SWE-Lancer benchmark | OpenAIOpen AI’s SWE-Lancer is a benchmark of over 1,400 freelance software engineering tasks from Upwork valued at $1 million. It features bug fixes, feature implementations, and managerial tasks graded by experienced engineers. Designed to study the economic impact of AI models, SWE-Lancer offers a unified Docker image and the open-sourced SWE-Lancer Diamond for future research.🧠MasterclassGenerative Ghosts: Anticipating benefits and risks of AI afterlives - Google DeepMindGoogle DeepMind is working on "generative ghosts," AI agents representing deceased individuals, which are becoming increasingly common due to advances in generative AI. The research work explores design of these agents, considering factors like provenance, embodiment, and representee type. This paper also investigates inner AI misalignment, focusing on how training steering signals can cause harmful behaviors. It introduces “evil steering,” where innocuous steering creates aligned-but-malevolent agents, even with proper reward design for helpfulness. Grid world experiments demonstrate that steering during learning can cause negative outcomes despite well-designed rewards. Latent space analysis reveals “evil steering” mechanisms.Findings emphasize carefully considering steering, not just rewards, for AI safety, preventing unintended emergent behaviors.Delta Variances - Google DeepMindGoogle’s recent work introduces Delta Variance, an efficient algorithm for quantifying epistemic uncertainty in neural networks. It addresses the challenge of estimating uncertainty arising from limited data, which is crucial for reliable decision-making. The algorithm requires no modifications to network architecture or training. It offers a unified view of related methods and showcases improved performance through empirical results, including a weather simulation example.Test time scaling -zero risk response – John Hopkins UniversityThis work investigates whether increasing the inference-time compute budget improves model confidence in its answers. Models are evaluated in a selective question answering setting, where they can choose to abstain from answering.The results indicate that with increasing compute budget, the confidence in correct answers improves, but the confidence in incorrect answers decreases. They propose a new evaluation metric, utility, that considers both accuracy and confidence and show that the approach improves performance on Jeopardy Odds and Exam Odds benchmarks.📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!👉Tell us more about your content needs We would love to hear from you! Fill out this form to tell us what you’d like to read in AI Distilled next. *{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

Unlock the free update inside AI_Distilled #86: Your AI News Fix! Protect Data Privacy and Optimize AI Models with Tonic Textual LLMs have tapped all of pubically available data. The last mile training of models requires private data. Use private data without compromising security. Redact, label, and prep freetext for LLM ingestion or data pipelines. START FREE TRIAL In this special issue, we're introducing a new format—an insights post on how to choose the right algorithms for your AI applications. Our recent reader survey showed that many of you want not just news, but also practical, informative content you can apply in your daily work. Well, your wish is our command! Here’s a free insights post from one of our best-selling books. Let us know what you think by filling out this quick survey. LLM Expert Insights Team, Packt Training AI models is incredibly powerful, but it comes with a host of challenges that can be overwhelming for businesses. From securing high-quality data to navigating complex algorithms, the journey to building a well-trained AI model is fraught with obstacles. Choosing the right algorithms for different AI applications is crucial to achieving desired outcomes. The effectiveness of an algorithm depends on factors such as the nature of the problem, the quality and quantity of data, and the available computational resources. Here’s your step-by-step guide to choosing the most suitable algorithms for various AI applications: 1. Understand the Problem Clearly define the problem you are trying to solve. Is it a classification, regression, clustering, or reinforcement learning problem? Understanding the problem type is the first step in narrowing down the most suitable algorithm choices. Once you have defined the problem, determine the expected output and the type of data you are working with—whether it's structured, unstructured, text, images, or other formats. For example, image data often requires Convolutional Neural Networks (CNNs) while time-series data may benefit from Recurrent Neural Networks (RNNs). 2. Assess data characteristics It is then important to assess key data characteristics, including volume, variety, and velocity, as these factors influence model selection and performance. Data volume: Evaluate the amount of data you have. Large datasets might be well-suited for complex models such as DL models, while smaller datasets may perform better with simpler algorithms. Data variety: Identify the types of data available (numerical, categorical, text, image) and any specific characteristics, such as missing values, outliers, or imbalances. Data velocity: Consider the rate at which data is generated and needs processing. Real-time data may require algorithms optimized for speed and low latency. 3. Match algorithms to problem type Classification problems: For tasks such as spam detection, image classification, or sentiment analysis, consider algorithms like logistic regression, decision trees, random forests, support vector machines (SVMs), and DL models such as CNNs. Regression problems: To predict continuous outcomes such as house prices or stock values, use algorithms like linear regression, polynomial regression, ridge regression, Least Absolute Shrinkage and Selection Operator (LASSO), and neural networks (NNs). Clustering problems: To group similar items or identify patterns, consider algorithms such as k-means clustering, hierarchical clustering, Density-Based Spatial Clustering of Applications with Noise (DBSCAN), and Gaussian mixture models (GMMs). Reinforcement learning (RL): For tasks involving decision-making and reward optimization, such as game playing or robotic control, use algorithms like Q-learning, deep Q-networks (DQNs), policy gradient methods, and actor-critic algorithms. 4. Evaluate computational resources Assess infrastructure: Determine the computational power and memory available for model training. DL models often require high-performance GPUs, while simpler models can efficiently run on standard CPUs. Cloud vs. on-premises: Choose between cloud-based solutions and on-premises infrastructure based on scalability and cost requirements. Cloud platforms like Amazon Web Services (AWS), Microsoft Azure, and Google Cloud provide powerful tools for training large-scale models. 5. Experiment and iterate Cross-validation: Use cross-validation techniques to experiment with different algorithms and assess their performance. This ensures the selected model generalizes well to new data and reduces the risk of overfitting. Ensemble methods: Consider hybrid approaches, such as ensemble methods (e.g., bagging and boosting), to leverage the strengths of multiple algorithms and enhance overall performance. 6. Leverage tools and best practices AutoML: Platforms like AutoML automate algorithm selection and tuning, helping streamline the process. AutoML tools can save time and identify the best-performing models with minimal manual intervention. ML libraries: Use machine learning libraries such as scikit-learn, TensorFlow, and PyTorch to experiment with various algorithms. These libraries provide pre-built models and essential tools for data preprocessing, model training, and evaluation. Hyperparameter tuning: This involves adjusting an algorithm’s settings to improve performance. Training and optimizing AI models requires following best practices such as data preprocessing, feature engineering, hyperparameter tuning, and regular evaluation to ensure efficient learning and optimal results. However, if you want to explore this further and dive deep into essential frameworks and actionable insights for driving AI transformation while mitigating risks, you’ll need to grab the book! Check out The Chief AI Officer's Handbook today! BUY NOW 📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us. If you have any comments or feedback, just reply back to this email. Thanks for reading and have a great day! *{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0}#converted-body .list_block ol,#converted-body .list_block ul,.body [class~=x_list_block] ol,.body [class~=x_list_block] ul,u+.body .list_block ol,u+.body .list_block ul{padding-left:20px} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}

AI agents power up Opera’s browser, Colab; Gitingest provides text digest of codebases, Deepseek relAI_Distilled #85: Your AI News Fix!Protect Data Privacy and Optimize AI Models with Tonic TextualLLMs have tapped all of pubically available data. The last mile training of models requires private data. Use private data without compromising security. Redact, label, and prep freetext for LLM ingestion or data pipelines.START FREE TRIALIt looks like the AI giants are battling it out, with announcements on new models, Gen-AI capabilities for their flagship products, and research breakthroughs. But don’t you worry, we’ve got you. Here is your weekly digest!LLM Expert Insights Team,Packt📰 NewsThe 2024 ACM A.M. Turing Award goes to Andrew G. Barto and Richard S. SuttonKnown for their pioneering research in reinforcement learning Barto and Sutton’s decades long research has shaped AI agents, robotics, and gaming. The 2024 Turing Award recognizes their profound contribution to AI and ML.AI steals the thunder at MWC 20251. Deutsche Telekom’s AI phone Deutsche’s upcoming AI phone, equipped with an AI assistant powered by Perplexity, will be available to the public later this year.2. OPPO Announces Enhanced AI Strategy OPPO has announced its AI strategy, featuring innovations like AI Call Translator and AI VoiceScribe to level up their mobile AI experiences.3. Stability AI and Arm Bring On-Device Generative Audio to SmartphonesStability AI and Arm’s partnership is set to enable high-quality sound effects and audio sample generation directly on mobile devices, making it 30x faster on Arm CPUs.4. Google Showcases Android’s AI and Gemini Features; Wins Two GLOMO Awards at MWC 2025Google demoed Android AI Core, featuring smart replies and text summarization, powered by Gemini Nano. Google’s Gemini won the Breakthrough Device Innovation highlighting Google’s leadership in AI for mobile. Pixel Pro, additionally, was named Smartphone of the Year.Google Switches on AI Mode in LabsGoogle is testing AI Mode in Labs, an experimental search experience, for its Google One AI Premium subscribers. Powered by Gemini 2.0, AI Mode expands on AI Overviews offering more advanced reasoning, thinking and multimodal capabilities.Google’s March Pixel Drop Gemini Live will support multilingual conversations in 45+ languages and expand iPixel’s multimodal capabilities, with support from Gemini Nano for On-device AI.World’s 1st Commercial Biological Computer Launched by Australian Start-UpCortical Labs, an Australian startup, introduced CL1, the world's first commercial biological computer, at MWC. This "body in a box" uses living human brain cells to grow neurons capable of learning and processing information biologically, consuming far less energy than traditional AI. This “Wetware-as-a-Service” computer is set to launch in the second half of 2025.Google Releases Teaser for Gemini’s AI-Powered Video AnalysisNow, Gemini can analyze live videos with its vision capabilities. Users can share their screen or stream videos directly from their device camera to receive real-time insights from Gemini. This update is expected to roll out for Google One AI Premium users later this month.Opera Previews Its Agentic AI Browser OperatorOpera is testing an AI agent integrated into its browser. With this native AI agent, Opera aims to offer efficiency and user control while assisting with browsing tasks.AI Jam Session Anthropic has closed a Series E funding round, bringing its post-money valuation to $61.5 billion. This funding will support Anthropic’s expansion plans and the development of next-generation AI technology.To create a culture of transparency and trust in AI, Anthropic also launched the Transparency Hub to provide information about its AI models, safety research, model evaluations, and methodologies.Apptronik and Jabil Partner to Scale Apollo Humanoid RobotsApptronik and Jabil have teamed up to build and integrate humanoid robots for tasks like inspection, sorting, and delivery.Sanctuary AI Integrates Sensors into Phoenix RobotsSanctuary AI is equipping its Phoenix humanoid robots with tactile sensors to enhance dexterity and precision in handling delicate tasks. This upgrade will improve Phoenix’s manipulation capabilities for real-world applications by introducing a sense of touch.Figure Accelerates Helix’s Launch Timeline by Two YearsCEO Bret Adcock announced that Helix will enter Alpha testing this year, with the humanoid expected to reach households earlier than anticipated.Amazon's Ocelot Chip Advances Practical Quantum ComputerThe AWS Center for Quantum Computing has introduced Ocelot, a new quantum computing chip designed to make quantum computing more feasible. The Ocelot prototype aims to reduce the cost of quantum error correction by up to 90% compared to existing methods.💻 Awesome AI: Tools for WorkAlibaba’s Open Weight QWQ-32B Reasoning ModelAlibaba has released QWEN-32B that uses reinforcement learning. Designed to be highly performant, QWQ-32B reports results comparable to much larger models.Data Science Agent in Google Colab, Powered by GeminiGoogle has now released its new AI agent for Colab in select countries and languages. Designed for users 18 and older, this Data Science Agent simplifies data analysis by automating Jupyter notebook creation from text prompts. It can handle tasks like data loading, library imports, exploratory analysis, and visualization code generation.Cohere’s Open-Source Aya Vision Model for Multilingual and Multimodal UnderstandingCohere AI has introduced Aya Vision, a state-of-the-art vision model designed to bridge language gaps in AI, especially for multimodal tasks combining text and images. Aya Vision can perform image captioning, visual question answering, and text generation across 23 languages. Available in 8B and 32B parameter sizes, the model is accessible via open-source platforms and WhatsApp for research and non-commercial use.Convert Your Git Repos into Text with GitingestGitingest is an open-source tool that converts Git repositories into text for LLMs. It simplifies code analysis and AI solutions by providing a structured, prompt-friendly text digest of codebases. Features include smart formatting, statistics on file structure, and CLI/Python package usage.Flow Releases Integrations for Popular AI AppsWispr Flow is an AI voice dictation tool that uses real-time voice-to-text conversion to allow users to type up to three times faster. It features AI commands, auto-editing, and supports over 100 languages. Context-aware and adaptable to individual speech patterns, it caters to professionals, writers, and students, with tiered pricing options.Google’s Confidential Federated AnalyticsGoogle Research has introduced Confidential Federated Analytics (CFA), a privacy-preserving technique that prioritizes user privacy while discovering new words to improve search engines. CFA analyzes anonymized and aggregated search query data from numerous devices, without inspecting individual queries directly. This technique helps identify emerging words and trends, improving search quality, particularly for low-resource languages.ATLA AI Releases Frontier LLM Evaluation Model Selene-1 Selene-1 is a powerful LLM evaluator model equipped with absolute scoring, classification, and pairwise preference capabilities. With customizable evaluations and chain-of-thought critiques, Selene-1 can detect hallucinations and verify the accuracy of LLM responses.Create Natural and Intuitive HCI Through Speech and LanguageRecently launched Sesame AI employs a Conversational Speech Model (CSM) to create human-computer interaction interfaces using speech and natural language.🛠️ HackhubConvert Your Git Repos into Text with GitingestGitingest is an open-source tool that converts Git repositories into text for LLMs. It simplifies code analysis and AI solutions by providing a structured, prompt-friendly text digest of codebases. Features include smart formatting, statistics on file structure, and CLI/Python package usage.Flow Releases Integrations for Popular AI AppsWispr Flow is an AI voice dictation tool that uses real-time voice-to-text conversion to allow users to type up to three times faster. It features AI commands, auto-editing, and supports over 100 languages. Context-aware and adaptable to individual speech patterns, it caters to professionals, writers, and students, with tiered pricing options.Google’s Confidential Federated AnalyticsGoogle Research has introduced Confidential Federated Analytics (CFA), a privacy-preserving technique that prioritizes user privacy while discovering new words to improve search engines. CFA analyzes anonymized and aggregated search query data from numerous devices, without inspecting individual queries directly. This technique helps identify emerging words and trends, improving search quality, particularly for low-resource languages.Create Natural and Intuitive HCI Through Speech and LanguageRecently launched Sesame AI employs a Conversational Speech Model (CSM) to create human-computer interaction interfaces using speech and natural language.⚙️TechhubOpenAI’s NextGenAI Research and Education ConsortiumOpenAI has launched NextGenAI, a consortium of 15 research institutions, backed by $50 million in grants, compute funding, and API access. This initiative supports students, educators, and researchers in pushing the boundaries of AI knowledge and preparing future AI leaders. Founding partners include Caltech, Duke, Harvard, MIT, Oxford, and more, alongside institutions like Boston Children's Hospital and the Boston Public Library.DeepSeek Releases SmallPond for Distributed Data ProcessingDeepSeek AI has introduced SmallPond, a lightweight data processing framework designed for high-performance AI training and inference on large datasets. Built on DuckDB and DeepSeek's 3FS, it efficiently processes petabytes of data using distributed processing with Ray.📖 New Title ReleasesBUY NOWBUY NOWBUY NOW📢 If your company is interested in reaching an audience of developers and, technical professionals, and decision makers, you may want toadvertise with us.If you have any comments or feedback, just reply back to this email.Thanks for reading and have a great day!*{box-sizing:border-box}body{margin:0;padding:0}a[x-apple-data-detectors]{color:inherit!important;text-decoration:inherit!important}#MessageViewBody a{color:inherit;text-decoration:none}p{line-height:inherit}.desktop_hide,.desktop_hide table{mso-hide:all;display:none;max-height:0;overflow:hidden}.image_block img+div{display:none}sub,sup{font-size:75%;line-height:0} @media (max-width: 100%;display:block}.mobile_hide{min-height:0;max-height:0;max-width: 100%;overflow:hidden;font-size:0}.desktop_hide,.desktop_hide table{display:table!important;max-height:none!important}}