Update: We are extending the MOSS workshop deadline to May 26th 4:59pm PDT (11:59pm UTC)

Ben Edelman@benedelman.bsky.social · last yr.

What if there were a workshop dedicated to *small-scale*, *reproducible* experiments? What if this were at ICML 2025? What if your submission (due May 22nd) could literally be a Jupyter notebook?? Pretty excited this is happening. Spread the word! sites.google.com/view/moss202...

What if there were a workshop dedicated to *small-scale*, *reproducible* experiments? What if this were at ICML 2025? What if your submission (due May 22nd) could literally be a Jupyter notebook?? Pretty excited this is happening. Spread the word! sites.google.com/view/moss202...

1/ Excited to share a new blog post from the U.S. AI Safety Institute! AI agents are becoming more capable, but they are vulnerable to prompt injections in external content – an agent may be given task A, but then be “hijacked” and perform malicious task B instead. www.nist.gov/news-events/...

Technical Blog: Strengthening AI Agent Hijacking Evaluations

Large AI models are increasingly used to power agentic systems, or “agents,” which can automate complex tasks on behalf of users.

nist.gov

For years, this mysterious undulating loop has lived at the top of my personal homepage.

I defended my PhD dissertation back in May. I didn't have time to share it widely then (newborn baby), but I think some of you might enjoy it, especially the opening chapters: benjaminedelman.com/assets/disse...

0/ I'd like to kick off my presence here with a question: why does learning work in practice? Why is the world such that we can we learn to predict things from other things in a computationally efficient way; why is "simplicity bias" empirically useful? Some explanations: