I am delighted to officially announce the National Deep Inference Fabric project, #NDIF .

ndif.us

NDIF is an U.S. National Science Foundation-supported computational infrastructure project to help YOU advance the science of large-scale AI.

In a new preprint with Jarek Blasiok, Rares Buhai, David Steurer, we show a surprisingly simple greedy algorithm that can list decode planted cliques in the semirandom model at k~sqrt n log^2 n --essentially optimal up to log^2 n. This ~resolves Jacob Steinhardt's open question.

Super excited to share that VisDiff has been accepted to #CVPR2024 and selected as an oral (90/11,532)! We will give a 15-min presentation going through the methods and exciting applications enabled by VisDiff. See you in Seattle!

Language models can imitate patterns in prompts. But this can lead them to reproduce inaccurate information if present in the context.

Our work (arxiv.org/abs/2307.09476) shows that when given incorrect demonstrations for classification tasks, models first compute the correct

Protein language models (pLMs) can give protein sequences likelihood scores, which are commonly used as a proxy for fitness in protein engineering. But what do likelihoods encode?

In a new paper (w/ Jacob Steinhardt) we find that pLM likelihoods have a strong species bias!

1/

Independent AI research should be valued and protected.

In an open letter signed by over a 100 researchers, journalists, and advocates, we explain how AI companies should support it going forward.

sites.mit.edu/ai-safe-harbor/

1/

Beating prediction markets with chatbots sounds cool. In a recent work arxiv.org/abs/2402.18563, we get somewhat close to that.

As another perspective, forecasting is a great capability domain to benchmark LM reasoning, calibration, pre-training knowledge, and more. 🧵1/n

Remember when Bing’s LLM Sydney threatened Marvin von Hagen for tweeting about its prompt?

Our paper shows how such unexpected behavior in LLMs emerges from feedback loops and provides recommendations for evaluation to capture feedback effects.

📰: arxiv.org/abs/2402.06627

1/

Accepted to oral #ICLR2024 !

*Interpreting CLIP's Image Representation via Text-Based Decomposition*

CLIP produces image representations that are useful for various downstream tasks. But what information is actually encoded in these representations?

[1/8]