Peter Henderson
@PeterHndrsn
Followers
5K
Following
2K
Media
241
Statuses
1K
Assistant Professor @ Princeton (RL+strategic decision-making+Law). Prev: Stanford (JD/PhD); McGill/Mila; Meta FAIR; Amazon; Cal Supreme Court.
NYC
Joined February 2012
Been seeing some discussion that rewards should/shouldn't be in the environment. But what if I told you RL is a big field with different perspectives. Here is a definition from RL, Bit By Bit! https://t.co/8oQupp9HzS h/t @Dilip_Arumugam
0
0
14
Very thankful for Princeton’s willingness to invest in compute thanks to hard work from many CS faculty. There’s still time to apply to Princeton for your PhD! We’ve got more compute on the way too! That being said, there definitely needs to be more investment in academic
Last night, @agupta and I hosted a great dinner with 14 professors at #NeurIPS2025 from leading academic labs across the US, and many cited compute in academia as "abhorrent". Out of curiosity I just pulled these stats. This is insane. To do meaningful AI research today you need
2
4
51
New paper: ReliabilityRAG — Effective and provably robust defense for RAG-based web search. A graph-theoretic approach for adversarial robustness! LLMs often search online for answers to questions. However, retrieved documents can be irrelevant, noisy, contradictory, or even
0
6
25
Appreciate the hard work of the ICLR PCs! They were put in a tough position but are clearly working around the clock to address issues. They deserve a lot of credit and downtime after this!
0
2
14
Excited to share our work and chat at #NeurIPS2025! DM or find me to say hi! Check out: LiveCodeBench Pro, dynamic risk assessments for offensive cyber agents, and a multimodal benchmark for oil & gas ad framing/greenwashing — plus talks and work at the BioSafe, RegML, and
0
4
25
I’ll be on a panel here at #NeurIPS soon, fresh off the plane. 🛬 🤖
Excited about our NeurIPS'25 tutorial Data Privacy, Memorization & Copyright in GenAI with Cooper (co-founder, GenLaw) & Joe (represents OpenAI, Stability in all US copyright litigations) We bring together ML researchers, with those who understand its legal implications. Pls RT
0
1
13
PLI: https://t.co/nxRYsdBsLe CITP: https://t.co/J7t4OVNKzF Or here for the cog sci track:
0
0
0
Want to do a post-doc at Princeton with me? Apply soon and mention my name either in the Princeton PLI post-doc program or the Princeton CITP fellows program. You can also shoot me an email after you submit an application. I'm particularly interested in folks looking to do work
1
3
20
PLI: https://t.co/nxRYsdBsLe CITP: https://t.co/J7t4OVNKzF Or here for the cog sci track:
0
0
0
Want to do a post-doc at Princeton with me? Apply soon and mention my name either in the Princeton PLI post-doc program or the Princeton CITP fellows program. You can also shoot me an email after you submit an application. I'm particularly interested in folks looking to do work
1
0
2
PLI: https://t.co/nxRYsdBsLe CITP: https://t.co/J7t4OVNKzF Or here for the cog sci track:
0
0
0
Want to do a post-doc at Princeton with me? Apply soon and mention my name either in the Princeton PLI post-doc program or the Princeton CITP fellows program. You can also shoot me an email after you submit an application. I'm particularly interested in folks looking to do work
2
3
27
PLI: https://t.co/nxRYsdBsLe CITP: https://t.co/J7t4OVNKzF Or here for the cog sci track:
0
0
0
Want to do a post-doc at Princeton with me? Apply soon and mention my name either in the Princeton PLI post-doc program or the Princeton CITP fellows program. You can also shoot me an email after you submit an application. I'm particularly interested in folks looking to do work
1
1
14
Make sure you apply here: https://t.co/76R3VZeYjj You can also fill our expression of interest here:
0
0
2
One last reminder before the application deadline, looking for RL PhD students to join my group at Princeton! Especially those interested in exploration, RL for open-ended discovery, or alignment! Application links below!
We're getting close to PhD application time. I'm looking for PhD students and post-docs to join our lab to work on reinforcement learning! If interested, make sure to mention my name in your statement and/or fill out the expression of interest form below!
2
11
16
Yup. Also tracks with a bunch of work looking into problems with momentum based and adaptive optimizers for TD learning and policy gradient methods (including some from us). Adaptive optimizers and momentum have biases in on-policy learning that create issues, so there are only
We had a similar result on some AlphaZero variants (Player of Games) we ran back in the day. Saw minimal improvement from anything other than SGD.
0
1
13
Significant economic gains from AI agents will likely rest on fixing prompt injections and adversarial attacks. Otherwise connecting an insecure system to private information is just asking for problems. Unfortunately that's among the most challenging areas of AI research.
Top of HackerNews today: our article on Google Antigravity exfiltrating .env variables via indirect prompt injection -- even when explicitly prohibited by user settings!
1
0
10