Vaibhav @vaibhav_p1234

No more XML prompting! Considering that Opus/Sonnet/Haiku are the best-in-class LLMs right now (seen a lot of people switch to claude from chatgpt in day to day use), there's been a huge demand for better tool use capabilities - build apps that can do web search, run code, and…

LlamaIndex 🦙 @llama_index

3 weeks ago

No more XML prompting! Considering that Opus/Sonnet/Haiku are the best-in-class LLMs right now (seen a lot of people switch to claude from chatgpt in day to day use), there's been a huge demand for better tool use capabilities - build apps that can do web search, run code, and…

3 59 240 87K 193

Download Image

🅰️ Anthropic Tool Calling 🛠️ @AnthropicAI just added tool use to their Claude 3 APIs, which is extremely useful for structured outputs. Here's a short video explaining how to use it from the latest langchain-anthropic package! @RLanceMartin also explains how to design…

Financial RAG eval just got spicier 🌶️ Cohere launched Command R+ today. I tested the LLM and it scored a 70.12% on financial RAG. That is the highest score I have seen on this eval to date. Excellent work from the @cohere team. Command R+ has over 100B parameters, so…

[CL] An Incomplete Loop: Deductive, Inductive, and Abductive Learning in Large Language Models E Liu, G Neubig, J Andreas [CMU & MIT] (2024) arxiv.org/abs/2404.03028 - Modern LMs can learn new tasks through instruction following, few-shot prompting, and instruction inference.…

keyboard_arrow_left Previous keyboard_arrow_right Next

The two hottest topics in AI right now are RAG and Agents. Every top company in the AI community is laser-focused on making AI Agents a reality. I'm working on a couple of YouTube videos to teach you how Agents work and how to build one. In the meantime, let me show you this:…

An underexplored area of optimizing RAG is figuring out how to dynamically allocate LLM compute based on the complexity of the question. 1️⃣ "What did the author do in college" or "What are the risk factors for Uber in 2021" - straightforward retrieval and synthesis 2️⃣ "Figure…

LlamaIndex 🦙 @llama_index

3 weeks ago

An underexplored area of optimizing RAG is figuring out how to dynamically allocate LLM compute based on the complexity of the question. 1️⃣ "What did the author do in college" or "What are the risk factors for Uber in 2021" - straightforward retrieval and synthesis 2️⃣ "Figure… https://t.co/0eNWvBz7KU

3 70 278 49K 249

Download Image

Over the last two days after my claim "long context will replace RAG", I have received quite a few criticisms (thanks and really appreciated!) and many of them stand a reasonable point. Here I have gathered the major counterargument, and try to address then one-by-one (feels like…

Yao Fu @Francis_YAO_

2 months ago

Over the last two days after my claim "long context will replace RAG", I have received quite a few criticisms (thanks and really appreciated!) and many of them stand a reasonable point. Here I have gathered the major counterargument, and try to address then one-by-one (feels like…

75 47 480 270K 242

The biggest alpha in coding with AI right now is: Debug with GPT-4, Code with Claude 3. GPT-4 is still king when it comes to logic, but it's extremely lazy. Meanwhile, Claude would do anything you ask. The duo together is unbeatable.

Boom! Open Source Perplexity is here and I have been testing for most of this day! It is amazing. ReadMultiplex.com how-to soon! “Build a Perplexity-Inspired Answer Engine Using Next.js, Groq, Mixtral, Langchain, OpenAI, Brave & Serper” Code: github.com/developersdige…

Build Advanced RAG over Insurance Documents without Hallucinations 📑 Insurance documents can oftentimes have complex tables on which claims are covered or not covered. We found that oftentimes naive RAG solutions (using standard parsers) aren't able to parse these tables -…

LlamaIndex 🦙 @llama_index

4 weeks ago

Build Advanced RAG over Insurance Documents without Hallucinations 📑 Insurance documents can oftentimes have complex tables on which claims are covered or not covered. We found that oftentimes naive RAG solutions (using standard parsers) aren't able to parse these tables -… https://t.co/X794GmDgg5

5 57 265 139K 288

Download Video

95% of Machine Learning solutions in the real world are for tabular data. Not LLMs, not transformers, not agents, not fancy stuff. Learning to do feature engineering and build tree-based models will open a ton of opportunities.

Andreessen Horowitz (@a16z) just published a report on how enterprises build Generative AI last week after speaking to dozens of Fortune 500 companies. This report is packed with interesting insights. Here are my top 6: 1. At the moment, OpenAI still has the dominant market…

LLM Agent Operating System Presents AIOS, an LLM agent operating system, which embeds large language model into operating systems repo: github.com/agiresearch/AI… abs: arxiv.org/abs/2403.16971

vLLM has blown every other LLM inference method out of the water. vLLM is an open-source library to serve large language models. It uses a new attention algorithm to deliver up to 24x higher throughput than HuggingFace Transformers without requiring model changes. This is a…

Simplify debugging with AI's ability to decipher complex error messages. Get clear, actionable solutions to keep your development workflow smooth and efficient. An essential tool for tight deadlines. #Debugging

We are rolling out Gemini 1.5 Pro API so that you can keep building amazing stuff on top of the model like we've seen in the past few weeks. Also, if you just want to play with Gemini 1.5, we removed the waitlist: aistudio.google.com Last, but not least, we pushed the model…

Jeff Dean (@🏡) @JeffDean

a month ago

We are rolling out Gemini 1.5 Pro API so that you can keep building amazing stuff on top of the model like we've seen in the past few weeks. Also, if you just want to play with Gemini 1.5, we removed the waitlist: aistudio.google.com Last, but not least, we pushed the model… https://t.co/EQ8VtKFiom

95 403 2K 607K 681

Excited to share that LLaVa-NeXT (also known as LLaVa-1.6) is now in @huggingface Transformers! Improves upon its predecessor, LLaVa-1.5, by incorporating * higher image resolutions * better OCR and reasoning capabilities * various LLMs (Mistral-7B, Yi-34B by @NousResearch) 1/2

My calls are being answered! Two new strong reward models (not DPO) added the day after release of RewardBench. 1. A mistral instruct 7B fine-tune (second overall!, SOTA 7b) huggingface.co/Ray2333/reward… -- just trained on a ton of datasets! 2. A gemma RM, from @weixiong_1 (running…

I've said it before, 2024 is the year of the synthetic dataset. Not just for fine-tuning but also for training from scratch. In her new guide the awesome @LoubnaBenAllal1 reveals the actual tips and tricks to: 1/ generate a good 25B tokens dataset AND 2/ train a good 1B LLM on…

LLMs like ChatGPT are an amazingly powerful breakthrough in AI and a transformative general purpose technology, like electricity or the internet. LLMs will reshape work and our lives this decade. They are not just a blurry photocopier or an extruder of meaningless word sequences.

keyboard_arrow_left Previous keyboard_arrow_right Next