Recently, an issue popped up on RewardBench like "is this broken, seems not very reproducible?" We did a deep dive into the numerical stability of open reward models, finding lots of things (blog link below): 1. DeBERTa’s batch inference on HuggingFace is not implemented, but

Yesterday Apple strongly advocated for on-device AI with Apple Intelligence! Today, we know they will run ~3B LLM on device (Mac, iPhone, iPad) using fine-tuned LoRA Adapters for different tasks, claiming to outperform other 7B and 3B LLMs! 👀 What we know: 📱 On-device 3B

New blog: Can LLM-based eval replace human evaluation? ehudreiter.com/2024/06/11/llm…

I gave a lecture at Stanford University CS 25. Lecture video: youtu.be/orDKvo8h71o?si… AI is moving so fast that it's hard to keep up. Instead of spending all our energy catching up with the latest development, we should study the change itself. First step is to identify and understand

These pop-out bezel animations in iOS 18 beta... could they be setting the stage for the upcoming iPhone 16 having solid state buttons? 🤔 These animations + haptics would then compensate for the feel of pressing physical buttons.

This was a fun paper

Heard about LM program optimization in DSPy but not sure how it works or where to start? Check out our new paper on *how* to optimize LM program instructions and demos! Packaged with our new SOTA LM program optimizer MIPROv2! Threads and notebooks coming later this week!

Interesting visualizations showing LLMs representing: - categorical concepts as simplices - hierarchical relations as orthogonality

Some cool/crazy lessons: - don't mention pink elephants if you don't want LLMs to think about them - the one by Charles 🎉 Frye of actually using bad data (labeled as such) is 🤯 Also: This is a nice, low-friction format for sharing one-off lessons!

I've been collecting all sorts of extra information about the new OpenAI structured outputs feature throughout the day I think my post here may now be the definitive guide to details you may have missed simonwillison.net/2024/Aug/6/ope…

RL has reached another local optimum. It keeps climbing. interconnects.ai/p/reverse-engi…