Victor TOTH’s Post

LLM's are the new transistors pt 2 Anthropic's recently released Claude 3 has a feature called DISPATCH_SUBAGENTS that allows it to invent prompts and recursively invoke itself (including multiple parallel copies) to solve problems... https://lnkd.in/dUk-2mVi So this is like using multiple transistors to amplify the signal coming out of another transistor. It still doesn't have feedback loops (while 'this is not a good answer to the prompt': try something else). Next we'll see feedback loops. Yann LeCun presents a pretty compelling biologically-inspired flowchart of what's coming next https://lnkd.in/dUEbshZD Also likely we won't be using discrete text tokens but a more general latent vector space, like OpenAI's Sora uses or V-JEPA. (Part of me likes the readability of pure text though :'( )

Have you heard of AI sub-agents? Simply put, it's a small trick that allows generative AI to take on large tasks. Without sub-agents, you might ask ChatGPT: "Please proof check this compliance report for my boss". With sub-agents, you might ask ChatGPT: "Please produce the compliance report that my boss requested". To achieve this, ChatGPT will break your task down into its component parts and carry them out: 1. Find the relevant files 2. Parse them 3. Aggregate the data 4. Proof check its own work etc. This requires significant time, computing power, and model intelligence, but these areas are all experiencing rapid improvements. Watch the 4 minute video below to see sub-agents in action. https://lnkd.in/e8ZAE6VS

8:8:8 LIONS GATE PORTAL explained. You are under surveillance of "Saturn x Sun" duo! Stay positive - talk positive for yourself, your family /friends and loved ones. Ensure you sync your desires with your actions and pure intention while manifesting today. PUNEET PRAKAS 🙏 #Numerology #DecodingNumbers #Numerologist #Data #Science #Numbers

🌟 RAG vs Context Window? Where do you stand on this debate? 🚀 🔍 Topic: "Why RAG Will Never Die - The Context Window Myth" 📅 March 27th, Time: 10:00 AM PST/ 6:00 PM CET 🌐 Registration: https://lnkd.in/gas4bMrW Join top experts, including our very own Amr Awadallah and Amin Ahmad as they dive into the resilience of RAG models and debunk myths around longer sequence lengths. #GenAI #RAG #ContextWindow

🌟 RAG vs Context Window? Where do you stand on this debate? 🚀 🔍 Topic: "Why RAG Will Never Die - The Context Window Myth" 📅 March 27th, Time: 10:00 AM PST/ 6:00 PM CET 🌐 Registration: https://lnkd.in/g6yyTpZc Join top experts, including our very own Amr Awadallah and Amin Ahmad as they dive into the resilience of RAG models and debunk myths around longer sequence lengths. #GenAI #RAG #ContextWindow

Check out latest work from our team that finds that Vision Language Models have strong blindspots that keep them from performing visual reasoning anywhere close to human accuracy. We tested how VLMs work on Raven's Progressive Matrices, which are a used extensively in IQ tests and found several failures modes that cause some of these failing. https://lnkd.in/g_b96XRp https://lnkd.in/gvsw4Z27 With coauthors - Yizhe Zhang, He Bai, Ruixiang ZHANG, Jiatao Gu, Shuangfei Zhai, Joshua Susskind

Deepfakes are being used to weaponize false information in the world of politics. This can have a significant effect on public opinion, and ideological leaning. The highly-realistic nature of deepfakes have made this a larger issue than ever before. Facia is working to combat this problem through our deepfake detection solutions. It is also important to be aware of the markers of a deepfake, including unnatural movement and digital artifacts. Stay protected from misinformation with Facia! #deepfake #technology #facia

Claude 3 is here. The team at Anthropic have introduced a new family of models that can outperform GPT-4 and Gemini while addressing common issues with production use cases like RAG. Here’s what you need to know ⬇️

Building a Multimodal RAG Pipeline with Claude 3 (Anthropic) and SingleStore A lot of data out in the wild is in the form of images, even within documents, and we see multimodal RAG playing a bigger and bigger role in the future. Here’s an article by Pavan Belagatti showing you all the basics on how to build a RAG pipeline over images - do structured extraction, and index the text with an embedding model into a vector db (singlestore). https://lnkd.in/gHU47jxf

🌟 RAG vs Context Window? Where do you stand on this debate? 🚀 🔍 Topic: "Why RAG Will Never Die - The Context Window Myth" 📅 March 27th, Time: 10:00 AM PST/ 6:00 PM CET 🌐 Registration: https://bit.ly/3VxmKL8 Join top experts, including our very own Amr Awadallah and Amin Ahmad as they dive into the resilience of RAG models and debunk myths around longer sequence lengths. #GenAI #RAG #ContextWindow