Paul Dysko’s Post

Matt berman does a quick review of LLAMA 3, a new opensource llm model from Meta AI right now chatGPT4 is the leader. Most ai app building tools are supporting it. I see some shift in that towards claude. i think it happens faster since openai is neither open nor cheaper to use. I dont see much value in Gemini use. It comes out way behind others on as an ai chatbot or even code generation. There is a wow factor that doesn't translate into hobbyist like me with simple computing needs. Groq and llama 3 will be a good foundation to build ai applications once ai tool builders start supporting this platform, compared to openai chatgpt4, a money sinkhole for running agent based ai applications..

Llama 3 is out! 🔠 Trained on 15T Tokens & fine-tuned on 10M human annotated samples 🧮 8B & 70B versions as Instruct and Base ✍🏻 Tiktoken-based tokenizer with a 128k vocabulary 🪟 8192 default context window (can be increased) 💰Commercial use allowed 🤗 Available on Hugging Face 🔜 more model sizes & enhanced performance

Hello LinkedIn community, I’d like to share my recent blog with Hasgeek - “Decoding Llama3.” 👋 It’s a deep dive into the Llama3 model code released in April this year. If you’re a technical person and want to understand the internals of the Llama3 model, then this is a fun blog with a code-first approach. I hope you find it interesting and please do share your feedback. https://lnkd.in/gmYdKXha

This is how you can train the open source AI "Llama 3" with your own company data, without sending anything over the Internet! 🦙 Llama 3 has the potential of being a turning point for medium- to large sized companies that wish to enroll their own LLM trained on their own company data. By purchasing a single A100 80GB GPU for ~25 000 Euro you can now download and train the free Llama 3 70B model (which will be on-par with GPT-4 once it has been fully trained) using tools like Unsloth with your own data in your own data facilities, without sending any information outside the company! Here are just a few use cases from the video: * Fine-tuning an LLM on customer service transcripts creates chatbots that can address issues in a way specific to your company * Tailored content generation: Fine-tune an LLM on your posts and descriptions to create engaging summaries or marketing copy. * Domain-specific analysis: Fine-tuning on legal or medical text enables LLMs to extract complex insights much faster than a human could. What an exciting time to be alive! 🌟 https://lnkd.in/djwcrbQB

It's out! Assaf Elovic walkthrough of how he used LangChain to build GPT Researcher (13k stars on github). Together we discuss the tradeoffs, complexities and advantages of running a multi agent system in production. A multi-agent framework is a system designed to support the development and operation of multiple autonomous software agents that can interact with each other and their environment. It typically includes mechanisms for agent communication, task distribution, decision-making, and coordination, enabling collaborative problem-solving. Pros of a multi agent system: 1. Separation of concerns 2. Scalable as systems grow 3. Optimal modularity 4. Reflects real world collaborative problem solving 5. System capabilities exceed the sum of individual agent capabilities (1+1 > 2) 6. Human-AI collaboration 7. Can handle non-linear workflows Youtube - https://lnkd.in/dvTqfyzK #GenAI #RAG #MultiAgent

The more I work with Large Language Models, the more I believe they are perfectly suited for assisting with important yet tedious documents (like regulations, manuals, policies, and procedures). As someone who tends to be on the lazy side, I appreciate any tool that helps me quickly zero in on the parts of these documents that matter for a specific problem. It makes me a happy (and still lazy) person! I’ve recently opened up my Exchange Control Tool (https://lnkd.in/dcXqPtP2) as an example of where I believe these tools truly shine—and why I think they’ll soon be everywhere. Imagine not having to train a new employee on all the legacy documentation and manuals they need to know, or not having to re-explain, for the hundredth time, how a process works to someone in another department. This future is just around the corner if you embrace generative AI. I've also spent some time getting the tool running in a more "production-like" environment. Now, anyone can use it without needing to sign up (although some corporate firewalls may still block access). Give it a try and let me know what you think!

I have had a few people ask me about GPTs recently so I figured I would reshare this video I published about a month ago. If you are a ChatGPT Plus user, you can build your own GPTs. But what does that mean? Watch this video for a crash course. If you have created one and want to share it with the world, post the link the comments! https://lnkd.in/eKSShSD2

Llama 3 by Meta is looking really good. It's already in the top-5 according to Large Model Systems Organization's side by side comparison testing. https://chat.lmsys.org/ I did a few blind tests comparing Llama 3 to GPT-4-Turbo. They both gave solid and accurate answers. One thing that stood out. Llama doesn't do the annoying preamble that ChatGPT does. The one where it overly reiterates how important the question you're asking is. https://lnkd.in/g7Gmtzve

Looking forward to testing it... 🚀 Llama 3 is the next iteration of Llama with a ~10% relative improvement to its predecessor! Llama 3 comes in 2 different sizes 8B and 70B with a new extended tokenizer and commercially permissive license! 𝗡𝗲𝘄 𝗮𝗻𝗱 𝗶𝗺𝗽𝗿𝗼𝘃𝗲𝗺𝗲𝗻𝘁𝘀 𝘁𝗼 𝘃𝟮: - Trained on 15T Tokens & fine-tuned on 10M human annotated samples - 8B & 70B versions as Instruct and Base - Llama 3 70B best open LLM on MMLU (> 80 ) - Instruct good at coding 8B with 62.2 and 70B 81.7 on Human Eval - Tiktoken-based tokenizer with a 128k vocabulary - 8192 default context window (can be increased) - Used SFT, PPO & DPO for alignment. - Commercial use allowed - Available on Hugging Face Blog: https://lnkd.in/gXrqAK6u Models: https://lnkd.in/gZSTfYED Chat Demo: https://lnkd.in/gF_zdBW9

Discover the Power of AI: Comparing Llama-3 and Phi-3 Using RAG 🚀 We are excited to share a fantastic resource recently released on Lightning AI's platform! This tutorial, titled "Compare Llama-3 and Phi-3 using RAG", offers an insightful and practical approach to understanding the capabilities and differences between two cutting-edge AI models, Llama-3 and Phi-3, through the Retriever-Augmented Generation (RAG) framework. 👉 What's Inside? The tutorial is a hands-on guide that walks you through the steps of setting up and running comparisons between Llama-3 and Phi-3. It provides a detailed overview of how each model performs in various tasks and how their results can be enhanced using the RAG technique. Whether you're a researcher, a data scientist, or simply an AI enthusiast, this tutorial is designed to enrich your understanding of modern AI technologies. 🔍 Why It Matters? Understanding the nuances between different AI models helps in choosing the right tool for specific applications, enhancing both efficiency and effectiveness in projects. The tutorial not only sheds light on the models' capabilities but also demonstrates the practical applications of these technologies in real-world scenarios. 🌐 Check out the full tutorial here: https://lnkd.in/ewZu3G_M