Congratulations to Avi Wigderson on winning this year's Association for Computing Machinery A.M. Turing Award! 🎉 His contributions to how randomness and computation interact have significantly shaped modern computing.

Big congratulations to Avi Wigderson of IAS Princeton for winning the Turing Award in CS. Truly an all-time great in theoretical computer science and discrete math. Also one of the nicest human beings I know --friend and mentor to so many (including me) tinyurl.com/fz5vxxaf

Very interesting papers Zeyuan Allen-Zhu . This trick is very interesting. I recall hearing evidence OpenAI does label training data with source/provenance (the LLM sometimes spits out those memorized labels). Can't remember where/who I learnt this from

Good data is key for all LLM work. This is a nice survey on how to do good data selection for LLMs (pertaining or fine-tuning). Kudos to Mengzhou Xia and Sadhika Malladi

Check out the Princeton Alumni Weekly profile of Andrés Monroy-Hernández ➡️ hci.social/@andresmh, who leads the Human-Computer Interaction Lab @princeton.

He's building tech that serves the greater good — including research aimed at making platforms like Uber and Doordash better for gig workers.

paw.princeton.edu/article/human-…

Fantastic work from our Princeton PLI team. Agent-based AI is clearly a big next step.

Yeah the grads here are amazing. Every day at work is a treat.

Delighted and honoured to receive a Knighthood for services to AI. It’s been an incredible journey so far building Google DeepMind over the past 15 years, helping accelerate the field and grow the UK & global AI ecosystems. Thanks to everyone who helped make this dream possible!

I was quite curious what OpenAI's preparedness unit is working on, and Aleksander Madry gave a good high-level view in our Princeton Alignment and Safety seminar youtube.com/watch?v=pe3XF0…
Kudos to Sadhika Malladi and Yangsibo Huang for interesting followup Q&A

Looking forward to understanding from Aleksander Madry about OpenAI 's thinking on near-term AI risks, responsibility of the preparedness unit. Tune in to the seminar tomorrow (Tuesday) 2pm EST!

Do Large Language Models (LLM) have leadership? Together with cognitive scientists Tom Griffiths Griffiths Computational Cognitive Science Lab and Natalia Velez Natalia Vélez (natvelali @ 🧵/🟦/fediscience.org) at Princeton University, we created a multi-LLM-agent system to test if agents work better with a leader. It turns out that LLMs do have leadership,…

Congratulations on the move Michal Valko

The GPUs have landed! (300 H100s on tap for AI research Princeton PLI )
ai.princeton.edu/news/2024/prin…

Very excited to host Paul Christiano as our kick-off speaker in Princeton Alignment and Safety Seminar today at 2pm pli.princeton.edu/events/2024/ca…

Tried it on Claude 3. It got it wrong badly, talking about arctic circle. Then I asked it 'How far do you walk from north pole to get to the arctic circle?' It immediately corrected itself and gave a (i.e., one of the) correct answer.

FWIW imagine the time and money (a few billion dollars at least) that they saved by just helping themselves to pirated books instead of trying to get permissions from a million authors.

Deepseek's new VLM tinyurl.com/yed42tpx is very impressive. But p 7 of mentions they trained on 1M books from 'Annas Archive', i.e., illegal downloads. That's 100B very high-quality tokens.  Dark new world...

Excited to kick off our Princeton Alignment and Safety Seminar with Paul Christiano on 3/19! Livestream link coming soon to the linked page tinyurl.com/pass-seminar

Excited to see how SWE Bench swebench.com from Princeton PLI group is now guiding use of AI for software engineering.

1/ Given high-dim time series data (e.g., video, text), we'd like to know 'what happens in the future?' and 'what happened between A and B?' Turns out temporal contrastive learning provably answers these questions!

Paper+Code: arxiv.org/abs/2403.04082
Backstory and details below