huggingface.co/papers/2307.02…

⚠️ Giveaway I Built a "Warren Buffett Investing Course" Step-by-step video course: - How to find high quality stocks - Value them - Determine an attractive price - and much more Valued at $159 FREE to 20 people, just: 1️⃣ Follow me 2️⃣ RT & Like 3️⃣ Reply 'Buffett' below

Open-source LLMs are now commonly used and widely studied, but this area of research saw some initial struggles and criticism due to the poor performance of models like OPT and BLOOM. These four recently-proposed open-source models changed this narrative… LLaMA. Interest in

Open challenges in LLM research The first two challenges, hallucinations and context learning, are probably the most talked about today. I’m the most excited about 3 (multimodality), 5 (new architecture), and 6 (GPU alternatives). Number 5 and number 6, new architectures and

More info on temperature and its inner workings can be found here. x.com/cwolferesearch…

There appears to be a mismatch between publishing criteria in AI conferences and "what actually works". It is easy to publish new mathematical constructs (e.g. new models, new layers, new modules, new losses), but as Apple's MM1 paper concludes: 1. Encoder Lesson: Image

Introducing `claude-journalist` ✍️ The first Claude 3 journalist agent. Just provide a topic, and it will: - Search the web for articles/real-time details - Choose the best sources and read through them - Write a fantastic, *factual* article + edit it And it's open-source!

A nice analogy of RAG vs. finetuning is to compare it to an open vs. closed-book exam 📖RAG == open-book exam without studying. Only use information provided in the page, but hard to discern which information is relevant. 📘Finetuning == closed-book exam. Only use memorized

Meet #DBRX: a general-purpose LLM that sets a new standard for efficient open source models. Use the DBRX model in your RAG apps or use the DBRX design to build your own custom LLMs and improve the quality of your GenAI applications. dbricks.co/43xaCMj

Introducing Jamba, our groundbreaking SSM-Transformer open model! As the first production-grade model based on Mamba architecture, Jamba achieves an unprecedented 3X throughput and fits 140K context on a single GPU. 🥂Meet Jamba ai21.com/jamba 🔨Build on Hugging Face

The Future Of Foundation Models? Directed Evolution Over Brute Force Imagine having the power to mix and combine the knowledge and capabilities of different AI models like mixing colors - taking the language understanding of one model, blending it with the math reasoning prowess

Last week, I described four design patterns for AI agentic workflows that I believe will drive significant progress this year: Reflection, Tool use, Planning and Multi-agent collaboration. Instead of having an LLM generate its final output directly, an agentic workflow prompts

I've been trying several evals to find those that correlate well with use cases and discriminative enough for prod. Here's an opinionated take on what works, focusing on classification, summarization, translation, copyright regurgitation, and toxicity. eugeneyan.com/writing/evals/

Prompt engineering is one of the most rapidly-evolving research topics in AI, but we can (roughly) group recent research on this topic into four categories… (1) Reasoning: Simple prompting techniques are effective for many problems, but more sophisticated strategies are

An open LLM that can be used for LLM-as-a-Judge evaluation as strong as OpenAI GPT-4 or Anthropic Claude 3? 🤯 Yes, KAIST AI just published PROMETHEUS 2, an open LLM specialized in evaluating other LLMs highly correlating with human and GPT-4 judgments. 🔥 Implementation:

Fixed continual finetuning in Unsloth AI! When continuing to finetune LoRA adapters, the loss goes haywire I accidentally set the tokenizer's padding side to "left", not "right" on new runs! Whoops! + save 20mins with multi GGUF options! Colab for both: colab.research.google.com/drive/1rU4kVb9…

We’ve found a 72% correlation between our clients ranking on the first page of Google and being mentioned by AI tools, such as ChatGPT and Perplexity. Here’s the approach we take for getting brands to show up in AI search. growandconvert.com/ai/google-seo-…

This paper studies vibe coding, a style where developers build apps by talking with an LLM instead of writing code. The authors conclude that skill stays vital, only the hands‑on parts move from keyboard to prompt and quick review. Many assume AI will erase coding grunt work

New Mistral AI cookbook: Self-Supervised Prompt Optimization ❌ Prompt engineering... sucks. It's a non-standard process, heavily relying on trial and error and difficult to standardize 🤩 Luckily, we can automate it using ✨prompt optimization✨, investigated in recent works

Gemini Nano Banana can be incredible if you know what prompts to use... i've built my own library of presets for ads, logos and all kinds of styles bookmark these 20 JSON templates: