Log inSign up
Zhili Feng
253 posts
user avatar
Zhili Feng
@zhilifeng
Research @OpenAI | prev: ML PhD @mldcmu | Research intern: @MSRNE, @Amazon, @BoschGlobal
Joined September 2010
1,036
Following
857
Followers
  • Pinned
    user avatar
    Zhili Feng
    @zhilifeng
    Jun 17, 2025
    πŸ₯³πŸ₯³πŸ₯³I defended my PhD thesis today! Special thanks to my wonderful advisor @zicokolter and committee members @rsalakhu @gneubig @LesterMackey! πŸŽ‰πŸŽ‰πŸŽ‰I am joining @OpenAI as a researcher, super excited to keep working on frontier models and meet everyone in SF!
    Image
    33K
  • user avatar
    Zhili Feng
    @zhilifeng
    Aug 19, 2024
    Last week I finished my internship @MSFTResearch, and I am very lucky to work with @LesterMackey @elmelis @nfusi. It is a truly memorable summer where I have learned so much and met so many amazing friends!✨✨
    Image
    Image
    Image
    4.6K
  • user avatar
    Zhili Feng
    @zhilifeng
    Sep 27, 2024
    Super excited to share that our work has been accepted to #NeurIPS2024! We advocate a new definition of data memorization from an adversarial prompt perspective. Thanks for the amazing collaborators @A_v_i__S @pratyushmaini @zacharylipton @zicokolter. Take a look at our paper:
    user avatar
    Pratyush Maini
    DatologyAI
    @pratyushmaini
    Apr 25, 2024
    Replying to @pratyushmaini
    2/ So what is Adversarial Compression? We argue that a given string from the training data is memorized if it can be elicited from an LLM by a prompt shorter than the string itself (AC Ratio > 1). i.e. it can be β€œcompressed” with the model via adv. prompts of fewer tokens.
    Image
    7.9K
  • user avatar
    Zhili Feng
    @zhilifeng
    Apr 30, 2025
    I'm very excited to talk about compression-based memorization with @pratyushmaini this Friday at the @OpenAI Security Research Conference! Let's chat about compression, memorization, and also our new antidistillation sampling antidistillation.com!
    3.4K
  • user avatar
    Zhili Feng
    @zhilifeng
    Aug 12, 2024
    This is a project that we are all very proud of! Thanks @pratyushmaini for presenting at CONDA@#ACL2024, and the great work of all the collaborators @A_v_i__S @zacharylipton @zicokolter πŸŽ‰πŸŽ‰
    user avatar
    Pratyush Maini
    DatologyAI
    @pratyushmaini
    Aug 10, 2024
    Our work on ACR memorization won the Best Paper Award at CONDA @ #ACL2024 πŸŽ‰πŸŽ‰ I will be giving a talk on the same on August 16th Over the last few months, I have substantially grown in the realization of how impactful ACR will be in the GenAI copyright discourse. Thoughts🧡1/n
    Image
    2.2K
  • user avatar
    Zhili Feng
    @zhilifeng
    Oct 9, 2024
    Now that Hopfield and Hinton won the Nobel Prize, Boltzmann machines seem to be interesting again, and I want to share this project I did with @ezra_winston and @zicokolter (who wrote most of the hard code) Monotone deep Boltzmann machines. (1/4) Paper: arxiv.org/pdf/2307.04990
    Image
    3K
  • user avatar
    Zhili Feng
    @zhilifeng
    Oct 21, 2024
    Compression for the win!πŸš€πŸš€πŸš€
    user avatar
    Yiding Jiang
    @yidingjiang
    Oct 21, 2024
    Selecting good pretraining data is crucial, but rarely economical. Introducing ADO, an online solution to data selection with minimal overhead. 🧡 1/n
    Image
    1.5K
  • user avatar
    Zhili Feng
    @zhilifeng
    Dec 9, 2024
    πŸ₯²Sadly I have to miss @NeurIPSConf, but my awesome collaborators will present our works at the main conference and a few workshops. I'm also looking for full-time research scientist or post-doc position, DM if you would like to chat!
    711
  • user avatar
    Zhili Feng
    @zhilifeng
    Oct 15, 2024
    Replying to @gautamcgoel
    I saw this arxiv.org/pdf/2103.00651
    15K
  • user avatar
    Zhili Feng
    @zhilifeng
    Oct 4, 2024
    I will be in Philadelphia #COLM from Oct 7-9, let's talk about unlearning/memorization, compression/mdl, Marvel movies, or anything! DMs open :) My amazing lab mates @pratyushmaini @A_v_i__S @_mingjiesun are coming as well, come find us!
    671
  • user avatar
    Zhili Feng
    @zhilifeng
    Jul 19, 2025
    Super exciting news and huge congrats to the team!! πŸš€πŸš€ One day when these models think long enough we may start to develop new math looking at their proofs, can’t wait to see that happens!
    user avatar
    Alexander Wei
    @alexwei_
    Jul 19, 2025
    1/N I’m excited to share that our latest @OpenAI experimental reasoning LLM has achieved a longstanding grand challenge in AI: gold medal-level performance on the world’s most prestigious math competitionβ€”the International Math Olympiad (IMO).
    Image
    956
  • user avatar
    Zhili Feng
    @zhilifeng
    Aug 21, 2024
    This particular puzzle seems underspecified: you can also pick the top-left pattern? πŸ€”πŸ€”πŸ€” @steph_milani
    user avatar
    ARC Prize
    @arcprize
    Aug 21, 2024
    How fast can you solve this one? Describe the solution in as few words as possible! ARC Prize Daily Puzzle arcprize.org/play
    ARC-AGI Task #39a8645d
    1.5K
  • user avatar
    Zhili Feng
    @zhilifeng
    Jan 25, 2024
    1/4 The privacy issue is getting more prevalent in the era of LLMs -- these models are trained on the crawl of the whole internet, what if an individual's data is inadvertently exposed to the training procedure? We introduce TOFU to benchmark machine unlearning process, a🧡
    Image
    483
  • user avatar
    Zhili Feng
    @zhilifeng
    May 6, 2024
    Wish I could be there tooπŸ₯²
    user avatar
    Pratyush Maini
    DatologyAI
    @pratyushmaini
    May 4, 2024
    TOFU was selected for an Oral PresentationπŸŽ™οΈat the Secure & Trustworthy LLM Workshop at ICLR2024. I will be talking about our new unlearning benchmark on behalf of my amazing co-authors @zhilifeng (who can't make it due to Visa issues πŸ₯²) & @A_v_i__S See you on May 11 at 10:30am
    Image
    330

New to X?

Sign up now to get your own personalized timeline!

Create account

By signing up, you agree to the Terms of Service and Privacy Policy, including Cookie Use.

TermsΒ·PrivacyΒ·CookiesΒ·AccessibilityΒ·Ads InfoΒ·Β© 2026 X Corp.
Don't miss what's happening
People on X are the first to know.
Log inSign up