🚀 Neal Lathia’s Post

Let's make RAG 40x faster and 32x memory efficient! To achieve this we'll be using binary quantization! 🚀 I have published this work as a Lightning AI ⚡️ Studio, It reads like a blog encapsulating all my code & environment! And here's what you'll learn: - Intuitive explanation of Binary Quantization (BQ) - Self-host Qdrant with BQ enabled - Search over 36M+ vectors in <50ms 🔥 Special thanks to my friend Nirant Kasliwal for his expert insights on how to use Qdrant to its fullest capacity! Everything you need right here👇 https://lnkd.in/gGMv5CAN

🦙🎉 We've expanded our Prompt Engineering with Llama 2 course to include the new Llama 3 model by AI at Meta! Interact with these models through API calls and: 🌟 Learn prompt engineering best practices for Llama 2 & 3 🧪 Experiment with few-shot and chain-of-thought prompting 💻 Code alongside Code Llama and learn to use Llama Guard to check prompts and responses Get started and let us know what you think of this update! Sign up here: https://hubs.la/Q02vH6kr0

📣 Pinecone's Roie Schwaber-Cohen's great article on Vectors + Graphs is now available on Neo4j's blog. 🎤 He introduces why Retrieval Augmented Generation (RAG) systems are important, and how vector + graph based systems make a powerful combination. 🎞️ Included are links to public notebooks showing how to implement such a system using Supreme Court Case data. The article is capped off with fantastic video showing a demo app he created to showcase and explore the composite results. ➡️ Read the full article at https://lnkd.in/gBEiaNzC

After a lengthy discussion with Perplexity Claude 3 Opus, I did not get the correct answer, so I asked again on GitHub Copilot and got the correct answer. I thought Opus was not as good as Copilot, but when I asked Opus the concise question, Opus also got the correct answer. Takeaway: Sometimes, start over with a new AI chat with a summarized and concise question. Why is this the case? No solution with a lengthy discussion: https://lnkd.in/grt8YQax A correct answer with a concise question: https://lnkd.in/grE46G2B

🤯 𝗧𝗵𝗮𝘁'𝘀 𝗿𝗲𝗮𝗹 𝗠𝗮𝗱𝗻𝗲𝘀𝘀 - 𝗮 𝗽𝗲𝗿𝘀𝗼𝗻𝗮𝗹 𝗔𝗜 𝗮𝘀𝘀𝗶𝘀𝘁𝗮𝗻𝘁 𝘄𝗶𝘁𝗵 𝘁𝗵𝗲 𝗳𝗮𝗰𝗲 𝗼𝗳 𝗘𝗺𝗺𝗮 𝗪𝗮𝘁𝘀𝗼𝗻. 🤖 One of the most impressive aspects of this AI assistant is its lightning-fast response time. It delivers answers in just one and a half seconds, thanks to a complex network of neurons, including wav2lip, Mistral, and Whisper, working tirelessly under the hood. 👉  The code is on GitHub (https://lnkd.in/ePqi9HPR). I love that it's open-source and accessible to everyone. Great move! 🙌 𝗦𝘁𝗮𝗿𝘁 𝗬𝗼𝘂𝗿 𝗣𝗿𝗼𝗷𝗲𝗰𝘁: 👉 𝗕𝗼𝗼𝗸 𝗮 𝗱𝗶𝘀𝗰𝗼𝘃𝗲𝗿𝘆 𝗰𝗮𝗹𝗹 𝗮𝘁 𝗵𝘁𝘁𝗽𝘀://𝗹𝗻𝗸𝗱.𝗶𝗻/𝗲𝗡𝟵𝗱𝟱𝗸𝘂𝗷

❓The big questions of AI Engineering❓ 🗃️ Should I use RAG or Fine-Tuning? ⚖️ 🪟 Will Long-Context windows kill RAG? 😵 ⛓️ LangChain or LlamaIndex? 🦙 What are your answers?  Drop your thoughts in the comments 👇 Join our upcoming Cohort 3 to discuss these questions and more with groups of like-minded peers out on the LLM edge!! Apply today! https://lnkd.in/g96iKq4f

Who better to give you a guided tour of CNTIENT∙Earth than one of the engineers involved in building it? Watch our Senior Software Engineer Izzy Hyman give a quick introduction to: 🧠 AI-Assisted Tasking: Easily creating and submitting single or recurring requests using natural language, and ☁ AI/ML-powered Feasibility Analysis: Reliable, ongoing order feasibility that understands the dynamic nature of the data providers’ supply chains and empowers satellite data consumers to make order decisions accordingly. Link to video in comments below because algorithm 👇 There's a lot more under the hood and many more features that we will share in upcoming product snippets, so keep an eye on this Page 👀

Is open sourcing crucial for AI tools in coding and testing? We've discussed this question with Eli Hooten, director of Engineering at Sentry. He says that LLMs are poised to benefit from open source contributions 👇 Full interview with Eli Hooten: https://lnkd.in/eCbwRAyU

We want to thank Amir Gholami, Research Scientist at #BAIR and Sky Lab at the University of California, Berkeley, for joining our lab to present his talk, "Efficient Function Calling Agents with LLMCompiler." If you're curious about the topic, here's the link to his publications and the abstract of his talk. Amir's pubs: https://lnkd.in/gc-JHMvR Abstract: LLMs are great at reasoning and taking actions. However previous frameworks for agentic reasoning (e.g. ReAct) were primarily focused on sequential reasoning, leading to higher latency/cost, and even poorer performance due to the lack of long-term planning. In this talk we will present LLMCompiler, a new framework that introduces a compiler for multi-function calling. Given a task, the framework plans out a DAG. This planning both allows for long-term thinking (which boosts performance) and determination of which steps can be massively parallelized. Benchmark results show that this approach allows consistent latency speedup of up to 3.7×, cost savings of up to 6.7×, and accuracy improvement of up to ~9% compared to ReAct, resulting in new SOTA accuracy results for AI Agents.

I'm done using ChatGPT to post on Linkedin. I know we should not have used it; LinkedIn penalizes AI-generated content. In my defense, I was not using chatGPT to create meaningless posts just to spam people's feeds. As soon as one of my articles on The Web Scraping Club went live, an automation built with Make summarized it and then published a post. Yes, I'm too lazy to write my summary of the articles. 😴 I have used this method for months, and everything seems fine until now. An example? Last Sunday, I published a wrap-up of the Extract Summit by Zyte. In the article, I clearly stated that I could not be there and commented on the speeches after I saw the videos on YouTube (in the comment in the original article). Well, ChatGPT summarized it, saying that it was a pleasure to attend the conference and speak in person with the web scraping community, cutting out all the meaningful content! Here's the link to the post: https://lnkd.in/dy2v-5Wr Okay, it's probably because I embedded some videos in the article, and this caused some issues with the summary, I said to myself. Yesterday, I wrote about Camoufox, a great tool for your #webscraping projects, to bypass the hardest anti-bot solution on the market (again, link in the comments). In the article, I wrote about the main contributor of the library several times and also reported some sentences from a conversation I had with him. Guess what GPT4 said? I was working on a new tool called Camoufox! This is totally false and misleading. I think it's time to roll up my sleeves and come back to the social networks game in first person! Have you had similar experiences with AI tools?