Working on the frontiers of AI safety? Excited about next generation AI safety challenges? Contribute to our #ICML2024 workshop on #NextGenAISafety! Submission: May 30th icml-nextgenaisafety.github.io Please reach out if you'd like to be a reviewer. Thanks for spreading the word!

Gemini 1.5 report is out. Checkout our safety section, as well! Naturally, I am biased, but I think it's by far the most detailed safety section in any report I have seen.

Submission deadline in 2 days!

Congratulations on the best paper runner up at trustnlp Yash Kumar Lal @ ACL2024 ! If you are at #NAACL2024, please swing by #TrustNLP workshop in Don Alberto 4 to catch the talk of our excellent intern!

Join us for a panel discussion on Building Trustworthy AI at the WiML Symposium at ICML Conference. Read more about our program here: buff.ly/4cYndf0

#crowdstrike BSOD appears to be turning into one of the worst global outages in the recent times. Major airlines, banks, broadcasters, and companies around the world offline...

I'll be at #ICML204 in Vienna, ping me if you are around! Beyond the corridors, you'll certainly find me at the following places:

Attending ICML 2024? Join us for an enlightening day of talks, mentorship sessions and networking at the WiML workshop on July 24th! Check out our program: buff.ly/4f3ix9y Let us know if you'll make it by filling the registration form: buff.ly/3WRW8oZ

Come to the poster and check out the paper if you'd like to learn more! arxiv.org/abs/2312.02592 This is joint work with amazing collaborators at Google DeepMind: Preethi Lahoti, Ben Packer, Yoni Halpern, Ahmad Beirami, and Flavien Prost.

WiML Panel Discussion: Building Trustworthy AI w/ Preethi Lahoti, Kamalika Chaudhuri, Emma Pierson, moderated by Luciana Benotti @ ICML Wed 24 Jul 2-2:45pm Schubert 1 - 6

Looking forward to talking about Safety and Fairness in LLMs at the Panel on "Building Trustworthy AI" at WiML. Do stop by Schubert 1-6 if you around at #ICML2024.

[#icml2024] Come meet Jong and I at poster #2712 at 11:30am to hear about controlled decoding from language models.

[#ICML2024] All accepted papers of the 𝐍𝐞𝐱𝐭𝐆𝐞𝐧𝐀𝐈𝐒𝐚𝐟𝐞𝐭𝐲 𝐰𝐨𝐫𝐤𝐬𝐡𝐨𝐩 are now publicly available on openreview. See you at the workshop tomorrow!

Happening now in Hall A1 at #ICML2024!

We classified harms into High Harm and Low Harm. Congratulations to Google DeepMind Gemini 1.5 Pro (post-IO) for being top of the list on both leaderboards!

Gemini 1.5 Pro is the safest model on the Scale Adversarial Robustness Leaderboard! We’ve made a number of innovations -- which importantly also led to improved helpfulness -- but the key is making safety a core priority for the entire team, not an afterthought. Read more about

Our team at Google is looking to host senior postdocs / junior faculty members in the areas of - language model alignment, - factuality and numerical accuracy, - human preference modeling and, - applications involving health and medicine Please feel free to reach out to me!