Princeton PLI
@PrincetonPLI
Followers
2K
Following
58
Media
61
Statuses
122
Princeton University initiative enhancing fundamental understanding of AI, enabling its use in academic disciplines, and examining AI's societal implications.
Princeton, NJ
Joined September 2023
How do LLMs think about thinking? π€ In the latest Kempner Seminar Series talk, @prfsanjeevarora (@PrincetonPLI, @PrincetonCS) explores how βmetacognitionβ shapes problem-solving in large language models. Watch here: https://t.co/SU7rlmXMnK Β #KempnerInstitute #AI #LLM #CS
0
2
9
Claude Skills shows performance benefits from leveraging LLM skill catalogs at inference time. Our previous work (linked under thread 5/5) showed the same 6 months ago! πOur new work, STAT, shows that leveraging skills during training can greatly help tooβΌοΈ, e.g., Qwen can
0
0
0
We welcomed 2 new associate research scholars this fall π₯³ Liam Fowl and @happybuzaaba1 will be supporting the research @PrincetonPLI to develop an understanding of LLMs and enable their application to research and education. Read more: https://t.co/KIPqsdTeAB
1
2
10
π¨ Call for Seed Grant Proposals π¨ The AI Lab is now reviewing seed grants to support research in large AI models, natural and artificial minds, AI for invention and other AI-related projects. Deadline for proposals is 10/31. For more information: https://t.co/xyF0adVSZp
0
2
9
EVERYONE PREFERS HUMAN WRITERS, INCLUDING AI We know humans are biased against AI-creativity but what about LLMs, now often judging creativity in various contexts? Check out new preprint from PLI Seed Grant Recipients, Meridith Martin and Wouter Haverals
0
1
1
Is online alignment the only path to go despite being slow and computationally expensive? Inspired by prospect theory, we provide a human-centric explanation for why online alignment (e.g. GRPO) outperforms offline alignment (e.g. DPO, KTO) and empirically show how to close the
4
32
172
0
2
9
What happens when AI is guided by law-like principles? Can we design some computational tools to "debug" rules? Check out our new work π ππ₯ππ₯π¦π₯π π£πͺ βπ ππ€π₯π£π¦ππ₯ππ π πππ πππ₯ππ£π‘π£ππ₯ππ₯ππ π ππ π£ πΈππ§ββοΈ to find out more! π§΅(1/10)
1
26
88
β±οΈAI is making verification process easier, with models verifying proofs in minutes. π» Now, @prfsanjeevarora, @chijinML, @danqi_chen and @PrincetonPLI have released Goedel Prover V2, a model more efficient and more accurate than any previous model. π https://t.co/v7500VNytz
1
21
94
Shoutout to all the @Princeton researchers participating in @icmlconf #ICML2025 Browse through some of the cutting edge research from AI Lab students, post-docs and faculty being presented this year: https://t.co/pCRjcKaH7G
0
9
48
Weβre proud that PLI students, post-docs, and faculty will be featuring over 20 papers at the @icmlconf in Vancouver this week! From safer AI agents to long-context reasoning and RL, weβre excited to showcase the cutting edge research for you here: https://t.co/invduoSPGJ
0
7
18
(1/4)π¨ Introducing Goedel-Prover V2 π¨ π₯π₯π₯ The strongest open-source theorem prover to date. π₯ #1 on PutnamBench: Solves 64 problemsβwith far less compute. π§ New SOTA on MiniF2F: * 32B model hits 90.4% at Pass@32, beating DeepSeek-Prover-V2-671Bβs 82.4%. * 8B > 671B: Our 8B
9
91
263
That's a wrap on the @PrincetonPIIRS Global Seminar in Kenya! Students discovered how to teach a computer to understand African languages & explored the current landscape of language technologies in Africa. Thanks to @happybuzaaba1 for this pic from the final poster session!
0
5
4
Are static evaluations enough to reflect the risk of offensive cyber security attacks? In our latest blog post, authors @wei_boyi @benediktstroebl @JiacenXu @JoieZhang @lzcarl and @PeterHndrsn show the answer is no! Read more:
pli.princeton.edu
@wei_boyi @benediktstroebl @JiacenXu @JoieZhang @lzcarl @PeterHndrsnMost frontier models today undergo some form of safety testing, including whether they can help adversaries launch costly cyberat...
2
6
15
π¦πQuACKπ¦π: new SOL mem-bound kernel library without a single line of CUDA C++ all straight in Python thanks to CuTe-DSL. On H100 with 3TB/s, it performs 33%-50% faster than highly optimized libraries like PyTorch's torch.compile and Liger. π€― With @tedzadouri and @tri_dao
13
76
342
βOur team of research software engineers have played a key role in the cutting-edge research in AI at Princeton that is being picked up by industry and also garnering awards and recognition at leading conferences.β Meet the AI Lab RSEs: https://t.co/2OUOkwBzhE
1
4
10
Yesterday the AI Lab hosted a year-end celebration to reflect on the work supported by our research initiatives and celebrate our collective achievements. Thanks to everyone who made this year a success β we look forward to the exciting path ahead!
0
1
8
Get to know the AI Lab's postdoctoral researchers with our new Q&A series, Meet the Postdoc First up, we're spotlighting @ethayarajh, who works on understanding how AI will operate as a part of human systems with @PrincetonPLI: https://t.co/f1QRP2XAj1
10
8
23
In a new blog post, @HowardYen1 and @xiye_nlp introduce HELMET andΒ LongProc, two benchmarks from a recent effort to build a holistic test suite for evaluating long-context LMs. Read now: https://t.co/8wfh1Qp2ES
0
11
19
In our blogβs new Deep Dive Series, @anilananth describes how a team at @PrincetonPLI, including @prfsanjeevarora, are looking to answer the questions: Do LLMs experience metacognition? And if so β how can it be leveraged to give models new skills? https://t.co/KVaJC3ieAE
1
2
3