user avatar
Adam Lerer
@adamlerer
Tuning hypers @AnthropicAI
San Francisco, CA
adamlerer.github.io
Joined February 2009
283
Following
3,336
Followers
Posts
  • Pinned
    user avatar
    Adam Lerer
    @adamlerer
    Nov 22, 2022
    1/ Today our paper describing a human-level AI for Diplomacy was published in Science (science.org/doi/10.1126/sc…)! This is the first human-level AI for a game requiring cooperation through *natural language*. Really proud of what we built and excited to finally share it.
    user avatar
    AI at Meta
    Meta
    @AIatMeta
    Nov 22, 2022
    Meta AI presents CICERO — the first AI to achieve human-level performance in Diplomacy, a strategy game which requires building trust, negotiating and cooperating with multiple players. Learn more about #CICERObyMetaAI: bit.ly/3GBwLzx
    00:00
    science.org
    Human-level play in the game of Diplomacy by combining language models with strategic reasoning
    Artificial intelligence demonstrates human-level performance in the strategic board game Diplomacy.
  • user avatar
    Adam Lerer
    @adamlerer
    Dec 6, 2020
    One of the more original papers I've had the pleasure to collaborate on, with @j_foerst and gang. We propose following the Hessian eigenvectors of the loss surface to find more diverse local minima.
    user avatar
    Jakob Foerster
    @j_foerst
    Dec 5, 2020
    The gradient is a locally greedy direction. Where do you get if you follow the eigenvectors of the Hessian instead? Our new paper, “Ridge Rider” (papers.nips.cc/paper/2020/fil…), explores how to do this and what happens in a variety of (toy) problems (if you dare to do so),.. Thread 1/N
  • user avatar
    Adam Lerer
    @adamlerer
    Oct 19, 2021
    Today we're releasing SIMBA, a tool for single-cell omics built on top of PyTorch-BigGraph. It’s been fun collaborating with @hd7chen, @lucapinello et al at the Broad to apply PyTorch-BigGraph, originally designed for large-scale web interaction data, to biology!
    user avatar
    Luca Pinello
    @lucapinello
    Oct 19, 2021
    Really proud of our new collaborative work with and @adamlerer at Facebook and led by the unstoppable @hd7chen and team. We present SIMBA a tool based on graph embedding to build gene regulatory maps.biorxiv.org/content/10.110…
  • user avatar
    Adam Lerer
    @adamlerer
    Oct 7, 2021
    We just released DORA, the first Diplomacy agent trained from scratch using RL+search without any human data. Joint work with @anton_bakhtin , David Wu and @polynoamial . arxiv.org/abs/2110.02924 So what?
    user avatar
    Noam Brown
    @polynoamial
    Oct 7, 2021
    Introducing DORA, an AI that learns no-press Diplomacy from scratch with no human data! Our #NeurIPS2021 paper shows DORA is superhuman in 1v1 Diplomacy. In 7p Diplomacy, the results are more subtle. Joint work w/ @anton_bakhtin, David Wu, and @adamlerer: arxiv.org/abs/2110.02924
  • user avatar
    Adam Lerer
    @adamlerer
    Feb 28, 2021
    A nice piece in IEEE Spectrum about our and DeepMind's work on Diplomacy. Finally, more research interest moving to large general-sum games!
    user avatar
    Matthew Hutson
    @SilverJacket
    Feb 26, 2021
    Forget Go, poker, and StarCraft. AI teaches itself Diplomacy. My latest for @IEEESpectrum, on work at @NeurIPSConf & @iclr_conf by @DeepMind & @FacebookAI. Thanks, @a_tacchetti, Yoram Bachrach, @adamlerer, @polynoamial. spectrum.ieee.org/tech-talk/robo…
  • user avatar
    Adam Lerer
    @adamlerer
    Sep 23, 2023
    Me and @alex_peys call this "humming": the LLM's gotta hum the tune to know the next line.
    user avatar
    Owain Evans
    @OwainEvans_UK
    Sep 22, 2023
    Does a language model trained on “A is B” generalize to “B is A”? E.g. When trained only on “George Washington was the first US president”, can models automatically answer “Who was the first US president?” Our new paper shows they cannot!
    13K
  • user avatar
    Adam Lerer
    @adamlerer
    Dec 7, 2020
    Go and Poker were both solved in 2017 but used quite different algorithms. I enjoyed working with @polynoamial and @anton_bakhtin on an AlphaZero-like approach that's sound for both perfect & imperfect info games. And using it to build a superhuman Poker bot!
    user avatar
    Noam Brown
    @polynoamial
    Dec 3, 2020
    Excited to announce our NeurIPS paper on ReBeL, an algorithm similar to AlphaZero that plays *imperfect-information* games like poker and Liar's Dice! Joint work with @anton_bakhtin, @adamlerer, Qucheng Gong. YouTube Video: youtube.com/watch?v=mCldyX… Paper: arxiv.org/abs/2007.13544
  • user avatar
    Adam Lerer
    @adamlerer
    Sep 7, 2023
    Replying to @hendrycks
    Based on my experience building and using Cicero, I think it's quite a stretch to say from these examples that "deception emerged as a subgoal". Any deception that emerged came from its training on Diplomacy dialogues, where deception is common.
    4.4K
  • user avatar
    Adam Lerer
    @adamlerer
    Apr 11, 2022
    It was a pleasure to work with @chloehsu0 during her internship last year! Our preprint is out describing our large scale inverse protein folding model, i.e. predicting sequence from 3D structure, trained on millions of sequences using AlphaFold2 predicted structures.
    user avatar
    Chloe Hsu
    @chloehsu0
    Apr 11, 2022
    Here’s what we learned from inverse folding on millions of #AlphaFold structures. Exciting time to bring a 800x new scale to #proteindesign. ESM-IF1 more accurately designs sequences to fold into desired structure, also unlocking new design capabilities. biorxiv.org/content/10.110…
    GIF
  • user avatar
    Adam Lerer
    @adamlerer
    Oct 13, 2021
    Pandemic lockdown gave me the opportunity to finally collaborate with @ZhongingAlong on two cryo-EM reconstruction papers, an exciting scientific area for protein structure determination! We just presented one on ab initio reconstruction @ICCV_2021! 1/
    user avatar
    Ellen Zhong
    @ZhongingAlong
    Oct 12, 2021
    CryoDRGN2 ❄️🐉 is now out at @ICCV_2021! We present an algorithm for *ab initio* reconstruction of heterogeneous protein structures. We focus on techniques for fast + accurate pose inference to achieve state-of-the-art reconstructions on all of our fav, real cryo-EM datasets 1/
  • user avatar
    Adam Lerer
    @adamlerer
    Nov 29, 2022
    Replying to @adamlerer
    If you have no clue what the game of Diplomacy is and/or want to learn how Cicero works, come to my talk this Friday 2pm at the Language and RL workshop @LaReL2022 at #NeurIPS22
  • user avatar
    Adam Lerer
    @adamlerer
    Nov 22, 2022
    Related: we put 2 CICEROs in a game with 5 top Diplomacy players just to see what would happen. One of them identified both AIs, one had no clue, and 3 accused their fellow players of being the bots :)
    user avatar
    Jerome Pesenti
    @an_open_mind
    Nov 22, 2022
    Replying to @an_open_mind and @metaai
    The most interesting anecdote I heard from the team: "during the tournament dozens of human players never even suspected they were playing against a bot even though we played dozens of games online."
  • user avatar
    Adam Lerer
    @adamlerer
    Nov 29, 2022
    Curious about CICERO in *anonymous human play*? Here’s a great game CICERO played where all 6 players allowed their dialogue to be published: dl.fbaipublicfiles.com/diplomacy_cice… Highlights: 🧵👇 (Links to all 40 games can be found at github.com/facebookresear…)
  • user avatar
    Adam Lerer
    @adamlerer
    Dec 6, 2022
    Cool new paper by @DeepMind on binding and non-binding agreements in Diplomacy.
    user avatar
    Google DeepMind
    @GoogleDeepMind
    Dec 6, 2022
    Today in @NatureComms, we explore how AI agents can better communicate and cooperate in Diplomacy - a 7-player board game of coordination and alliance formation. 🤝 Find out more: dpmd.ai/diplomacy-natu…
    GIF