We are hiring research scientists at Robust Intelligence Robust Intelligence. If you are interested and also attending ICML Conference please find me (or pm me) so that we can chat. I will be around from Monday until Saturday.

#ISMP2024 off to a good start!

Congratulations to the one and only Steve Wright! You likely know Steve from his scientific contributions and great books. And if you're extra lucky and have met him, you've also experienced firsthand what an amazing person he is. Cheers to one of the *absolutely* greatest!

New VLM reasoningEval Benchmark by Ahmad Beirami Ananda Theertha Suresh #ICML2024

We’re hiring researcher engineers at Databricks Mosaic Research! Interested and think you might be a good fit? DM me!

your nearest neighbor disagrees

This short report proves that a small sized one layer transformer can not perform induction head task arxiv.org/abs/2408.14332

Very cool paper

this is cool. Useful to remember though that the actual game is less than 3MB and the diffusion model is probably a few GBs

kinda painful how many OOMs are lost on the path from special to general purposes

This semester, I’m back to live blogging my course lectures. I’m teaching Convex Optimization in the Age of LLMs. argmin.net/p/convex-optim…

interesting paper with counterintuitive findings: Synthetic data from cheaper/weaker models can lead to better finetuning results in comparison to sampling from a more expensive/capable model, as you can sample way more at the same price As long as there's signal that is :)

The UW-Madison philosophy department is aiming for 3-6 hires this year, in any area, both junior and senior lines available! jobs.wisc.edu/jobs/professor… jobs.wisc.edu/jobs/professor…

One of the big problems in AI is that the systems often hallucinate. What does that mean exactly and how do we mitigate this problem, especially for RAG systems? 1. Hallucinations and Factuality Factuality refers to the quality of being based on generally accepted facts. For

Excited to share pre-print on AutoGen Studio! 🤖🤖🤖 "We present #AutoGen Studio, a no-code developer tool for rapidly prototyping, debugging, and evaluating multi-agent workflows built upon the AutoGen framework." Paper: arxiv.org/abs/2408.15247 Code: github.com/microsoft/auto…