Rajat Sahay
LightNews LightNews
rajatsahay.bsky.social
Rajat Sahay
@rajatsahay.bsky.social
230 followers 130 following 3 posts
ELLIS PhD | Interpreting world models
Posts Replies Media Videos
rajatsahay.bsky.social
Consider submitting to our workshop at #CVPR2025
sukrutrao.bsky.social Sukrut Rao @sukrutrao.bsky.social · Mar 2
Submit your latest work (papers, demos) in #XAI to the 4th Explainable AI for Computer Vision (XAI4CV) Workshop at #CVPR2025!

Details: xai4cv.github.io/workshop_cvp...
Submission Site: cmt3.research.microsoft.com/XAI4CV2025

@cvprconference.bsky.social @xai-research.bsky.social
Call for papers for the XAI4CV Workshop at CVPR 2025. Submission deadline on March 10, 2025 for the Proceedings Track.
March 3, 2025 at 11:29 AM
Reposted by Rajat Sahay
yoshuabengio.bsky.social
Today, we are publishing the first-ever International AI Safety Report, backed by 30 countries and the OECD, UN, and EU.

It summarises the state of the science on AI capabilities and risks, and how to mitigate those risks. 🧵

Full Report: assets.publishing.service.gov.uk/media/679a0c...

1/21
January 29, 2025 at 1:50 PM
Reposted by Rajat Sahay
bcmerchant.bsky.social
OpenAI responded to Musk's lawsuit, releasing emails that show it discussed becoming a for-profit as early as 2017.

As I showed in my report, being a nonprofit org was crucial to its mythology, recruitment, and eventual valuation—and it was performative from the start.

openai.com/index/elon-m...
December 13, 2024 at 8:43 PM
Reposted by Rajat Sahay
jparkerholder.bsky.social
Introducing 🧞Genie 2 🧞 - our most capable large-scale foundation world model, which can generate a diverse array of consistent worlds, playable for up to a minute. We believe Genie 2 could unlock the next wave of capabilities for embodied agents 🧠.
December 4, 2024 at 4:01 PM
Reposted by Rajat Sahay
phillipisola.bsky.social
Sharing some new work!

A big dream in AI is to create world models of sufficient quality that you can train agents within them.

Classic simulators lack visual diversity and realism. GenAI lacks physical accuracy. But combining the two can work pretty well!

Paper: arxiv.org/abs/2411.00083
November 14, 2024 at 1:30 AM
Reposted by Rajat Sahay
abursuc.bsky.social
The return of the Autoregressive Image Model: AIMv2 now going multimodal.
Excellent work by @alaaelnouby.bsky.social & team with code and checkpoints already up:

arxiv.org/abs/2411.14402
November 22, 2024 at 9:44 AM
Reposted by Rajat Sahay
adam-lg.bsky.social
"Do causal predictors generalize better to new domains?" -- Seems like a bit of a bombshell if you do causal work!

Paper: arxiv.org/abs/2402.09891

#stats 📉📈
We study how well machine learning models trained on causal features generalize across domains. We consider 16 prediction tasks on tabular datasets covering applications in health, employment, education, social benefits, and politics. Each dataset comes with multiple domains, allowing us to test how well a model trained in one domain performs in another. For each prediction task, we select features that have a causal influence on the target of prediction. Our goal is to test the hypothesis that models trained on causal features generalize better across domains. Without exception, we find that predictors using all available features, regardless of causality, have better in-domain and out-of-domain accuracy than predictors using causal features. Moreover, even the absolute drop in accuracy from one domain to the other is no better for causal predictors than for models that use all features. In addition, we show that recent causal machine learning methods for domain generalization do not perform better in our evaluation than standard predictors trained on the set of causal features. Likewise, causal discovery algorithms either fail to run or select causal variables that perform no better than our selection. Extensive robustness checks confirm that our findings are stable under variable misclassification.
October 24, 2024 at 2:25 AM