After 3+ years, today is the day that my book “The Worlds I See” gets to see the world itself. It is a science memoir of the intertwining histories of me becoming an #AI scientist, and the making of the modern AI itself. All versions are now on Amazon a.co/d/fYKf74L 1/

I completely missed ICLR blogpost track. It's a great idea that could benefit from some more visibility. iclr-blogposts.github.io/2024/call/

In Dec 2022, we released HELM for evaluating language models. Now, we are releasing HEIM for text-to-image models, building on the HELM infrastructure. We're excited to do more in the multimodal space!

📢Excited to share our new NEJM AI paper studying clinical adoption of #AI using billions of insurance claims! onepub-media.nejmgroup-production.org/ai/media/b35da…? We found that most US medical AI claims come from a few AI models (eg diabetic retinopathy), which are more likely to be used near academic

The goal is simple: a robust, scalable, easy-to-use, and blazing fast endpoint for open models like LLama 2, Mistral, etc. The implementation is anything but. Super impressed with the team for making this happen! And we're not done yet...if you're interested, come talk to us.

I've been using a GPT4 paper assistant that reads the daily ArXiv feed and makes personalized recommendations in Slack. It's worked pretty well for me (today's paper demo tatsu-lab.github.io/gpt_paper_assi…). If this sounds helpful, you can set up your own bot here github.com/tatsu-lab/gpt_….

Finally, we have a hallucination leaderboard! 😍😍 Key Takeaways 📍 Not surprisingly, GPT-4 is the lowest. 📍 Open source LLama 2 70 is pretty competitive! 📍 Google's models are the lowest. Again, this is not surprising given that the #1 reason Bard is not usable is its

Exited to introduce our new method as an alternative to #LLM in-context learning by latent space steering.

#ICLR2024 ICLR 2026 score statistics (7304 papers): mean: 5.10; max: 8.67; min: 1.00 >8.5: top 0.59-0.67% 8.0: top 0.68-1.75% 7.5: top 1.93-3.35% 7.0: top 5.20-8.17% 6.75: top 8.42-10.34% 6.5: top 12.13-15.39% 6.25: top 17.48-20.56% 6.0: top 21.02-28.29% 5.75: top 29.07-33.27%

Mara Graziani Accepted papers at TMLR Nice work! You might be interested about this one where we use NMF instead. arxiv.org/abs/2211.10154

We are presenting our poster on links between GANs and Diffusion at Neurips. Don't be scared by the formalism, ⁦Jean-Yves Franceschi⁩ is awesome and will explain everything intuitively!

🥳🥳🥳Our work on algorithmic fairness generalization has been accepted to #ICLR2024! TRAIL - Trustworthy and Responsible AI Lab Written with Grari Vincent Tatsunori Hashimoto Sylvain Lamprier Marcin Detyniecki during my visiting stay CodeX Stanford ❤️ Paper: arxiv.org/abs/2310.18413 Some details below 👇

💡Group fairness algorithms may generalize poorly: although a model may seem "fair" globally and appear to provide the same opportunities to men and women overall, this may not hold when looking at a subpopulation of the dataset, e.g. people over 60.

Does #RAG/web search solve #LLM hallucinations? We find that even with RAG, 45% of responses by #GPT4 to medical queries are not fully supported by retrieved URLs. The problem is much worse for GPT-4 w/o RAG, #Gemini and #Claude arxiv.org/pdf/2402.02008… RAG ≠ faithful to source

Important paper from Chelsea Finn and Sergey Levine lab: arxiv.org/abs/2402.19432… The authors trained a unique diffusion policy on a real-world dataset of various robots and tasks (manipulation arms, wheeled robots, self-driving cars, robot dogs, drones). Co-training gets 5%-20%

Levanter has the Sophia optimizer now, so you can train models ~2x faster. Together with Llama + Mistral, LoRA, TPU + GPU support, reproducibility, scalability, legibility, clean codebase, why not give Levanter a spin for your next LM training/fine-tuning run?

# CUDA/C++ origins of Deep Learning Fun fact many people might have heard about the ImageNet / AlexNet moment of 2012, and the deep learning revolution it started. en.wikipedia.org/wiki/AlexNet What's maybe a bit less known is that the code backing this winning submission to the

Attending #ICLR2024 ? Join Grari Vincent and I this afternoon at 4:30pm for our poster session! In our work, we propose a new method, ROAD, to build models that are not only fair globally, but locally as well!