🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Today was my last day at Lacework, after an eventful 2.5 years working on cloud security. I was fortunate to have excellent engineers as colleagues and learned a lot. Now ... it's time for a new adventure (following a short break).

Some news. I've joined AI at Meta (FAIR) to work on Reasoning, for code and other-than-code. And doubtless more of the great chewey problems one finds in AI these days. Gonna be fun!

Top 3 most cited AI papers of 2023, Llama, Llama2, SegmentAnything. All from AI at Meta

Our paper "Non-termination Proving at Scale", joint work with Azalea Raad and Julien Vanegue, uses compositionality to apply non-termination proving (classic undecidable problem) to large codebases (e.g., OpenSSL), finding non-termination bugs in the wild dl.acm.org/doi/pdf/10.114…

On my way to Pasadena for OOPSLA. Come check me out tomorrow at 2:40 to hear about non-termination proving at scale. I am recruiting for our CTO infra & security research team: full time, consulting, internships. I love meeting new creative people! Don’t be shy & hit me up!

Very proud of Noam Zilberstein! 🎉🎉 He’s doing amazing work and will go on the job market next year 👀 stay tuned for some of his new work appearing soon at POPL!

Lots of sense in these slides about trying to make formal methods practical, including lessons about things that have been challenging, assumptions we might question. And where the field might go to address the challenges.

🚨 New intern position in the FAIR CodeGen Team! 🚨 I'm particularly interested in working with candidates with expertise in off-policy RL methods, and/or in code generation with LLMs, but this is not a hard requirement and the project topic is somewhat flexible.

"AGI" meets Monty Python. Entertaining stuff!

Excited about our new ICLR workshop on AI + Verification! In the age of increasingly capable models, trusting outputs and getting high-quality feedback to improve models are becoming central bottlenecks. Our workshop explores how AI can be combined with formal systems (e.g.

Interesting chat about reasoning versus retrieval, centering on a concept of "procedural knowledge" and how it interacts with "influence" differently than non-procedural knowledge (basic fact retrieval) does.

Very high quality and influential work on reasoning about concurrent programs. Fully deserved POPL MIP Award.

And now for something (not) completely different: researching "reasoning" in AI; great fun is in store. ucl.ac.uk/computer-scien…

Generating tests that matter. Ask llm to make s mutant program with a new bug (e.g. introduce privacy violation). Ask llm to find test to squash bug but which original program passes: test is not useless, protects against introducing bug. Cool work! engineering.fb.com/2025/02/05/sec…

Harden and Catch for Just-in-Time Assured LLM-Based Software Testing: Open Research Challenges arxiv.org/abs/2504.16472 Paper with Mark Harman and Shubho Sengupta to go with our keynote at the Foundations of Software Engineering (FSE) 2025 conference in June.

The cat is out of the bag! Our very first Bloomberg Infrastructure & Security Research Ph.D. Fellowship is just announced. Consider submitting your fellowship application by July 18th.