Extremely excited to have this work out, the first paper from the Superalignment team! We study how large models can generalize from supervision of much weaker models. twitter.com/OpenAI/status/…

Excited to give an invited talk tomorrow Sep 30 at the #ECCV workshop on Uncertainty Quantification in Computer Vision at 12:25pm CET in room Brown3! I will present our research on spurious correlations and geographic biases in large-scale vision and multi-modal models!

Don't miss our tutorial A Bayesian Odyssey in Uncertainty: from Theoretical Foundations to Real-World Applications @eccvconf on Mon AM (08:45-13:00) #ECCV2024 In the line-up we have: @Pavel_Izmailov @GianniFranchi10 @a1mmer @_olivierlaurent @alafage_ uqtutorial.github.io

Modern generative models are trained to imitate human experts, but can they actually beat those experts? Our new paper uses imitative chess agents to explore when a model can "transcend" its training distribution and outperform every human it's trained on. arxiv.org/abs/2406.11741

Whether LLMs can reliably be used for decision making and benefit society depends on whether they can reliably represent uncertainty over the correctness of their outputs. There's anything but consensus. In new work we find LLMs must be taught to know what they don't know. 1/6

Excited to share what I've been working on as part of the former Superalignment team! We introduce a SOTA training stack for SAEs. To demonstrate that our methods scale, we train a 16M latent SAE on GPT-4. Because MSE/L0 is not the final goal, we also introduce new SAE metrics.

I'm excited to join @AnthropicAI to continue the superalignment mission! My new team will work on scalable oversight, weak-to-strong generalization, and automated alignment research. If you're interested in joining, my dms are open.

I’m excited to announce that I’ll start as an assistant professor at Columbia University this summer! Interview season was fun, I met so many amazing people, but I’m happy to finally close the loop.

Please submit your work on robustness / privacy / trustworthiness / alignment / ... with multimodal foundation models to our ICML workshop!

New Anthropic research paper: Scaling Monosemanticity. The first ever detailed look inside a leading large language model. Read the blog post here: anthropic.com/research/mappi…

An image is worth more than one caption! In our #ICML2024 paper “Modeling Caption Diversity in Vision-Language Pretraining” we explicitly bake in that observation in our VLM called Llip and condition the visual representations on the latent context. arxiv.org/abs/2405.00740 🧵1/6

📢Workshop on Reliable and Responsible Foundation Models will happen today (8:50am - 5:00pm). Join us at #ICLR2024 room Halle A 3 for a wonderful lineup of speakers, along with 63 amazing posters and 4 contributed talks! Schedule: iclr-r2fm.github.io/#program.

As part of our commitment to open science, I'm excited to share that an alpha version of our protein design code is available! Check out the tutorials to learn how to designs proteins yourself with guided discrete diffusion, just like the pros github.com/prescient-desi…

Excited to share our #ICLR2024 work led by @megan_richards_ on geographical fairness of vision models! 🌍 We show that even the SOTA vision models have large disparities in accuracy between different geographic regions. openreview.net/pdf?id=rhaQbS3…

If you're coming to @eccvconf consider our freshly accepted tutorial on: A Bayesian Odyssey in Uncertainty: from Theoretical Foundations 📝 to Real-World Applications 🚀 w/ the amazing @GianniFranchi10 @_olivierlaurent @a1mmer @Pavel_Izmailov More info coming soon #eccv2024

Another fascinating discussion this morning on the future of generative AI in physics, including the insight that "we'll be like cats...is that so bad?" (But in all seriousness, a lot of compelling back and forth on what it means to do science).

We are excited to be organizing a Symposium on the Impact of Generative AI in the Physical Sciences next Thursday, March 14 and Friday, March 15! Join us on the 8th Floor of @MIT_SCC for a great lineup of speakers and panelists. Zoom link available soon. iaifi.org/generative-ai-…

I'm excited to be speaking tomorrow at Boston University, as part of their distinguished speaker series. My talk will be on prescriptive foundations for building autonomous intelligent systems. Talk details: bu.edu/hic/air-distin…

I was very impressed by @martinmarek1999 in this project, look out for more exciting research from him!

Have you ever done a dense grid search over neural network hyperparameters? Like a *really dense* grid search? It looks like this (!!). Blueish colors correspond to hyperparameters for which training converges, redish colors to hyperparameters for which training diverges.

I'm glad to see losslandscape.com is still going strong. @ideami has beautiful visualizations. The geometric properties of neural network training objectives, such as mode connectivity, make deep learning truly distinct.