Clément Dumas @Butanium_
Astra fellow w/ Owain Evans ex MATS 7/7.1 Scholar w/ Neel Nanda and intern at DLAB (EPFL) AI safety research / improv theater butanium.github.io Joined December 2018-
Tweets1K
-
Followers1K
-
Following693
-
Likes19K
@wesg52 oh yeah I'm a big fan of this section too!!
Low hanging fruit (although §7 is kinda diffing!)
Don't miss those appendix in the new Global workspace anthropic paper: A.9: Extending the Jacobian lens to multi-token concepts A.21 Measuring evaluation awareness with the J-lens
@voooooogel If you haven't seen it, check out appendix A.9 where we explore some extensions that circumvent the single-token constraint!
@Jack_W_Lindsey @neuronpedia indeed resetting the zoom fixed the problem, thx!
I thought this was an excellent paper! Thanks to Anthropic for asking me to write a review of it, linked below I've long suspected that models have some kind of "working memory" to store intermediate variables during a forward pass and IMO this paper has the best evidence yet
New Anthropic research: A global workspace in language models. Of everything happening in your brain right now, only a tiny fraction is consciously accessible—thoughts you can describe, hold in mind, and reason with. We found a strikingly similar divide inside Claude.
Eleos wrote a commentary Tl;dr -important, excellent work -we’re more cautious than the authors about the stronger claims of 'global workspace' -still, it's evidence in the direction of access consciousness -investigating AI consciousness is tractable and urgent More:
New Anthropic research: A global workspace in language models. Of everything happening in your brain right now, only a tiny fraction is consciously accessible—thoughts you can describe, hold in mind, and reason with. We found a strikingly similar divide inside Claude.
@Jack_W_Lindsey @neuronpedia Very cool!!! The webpage is very laggy though :( Also has some rendering artifacts with paragraphs moving up and down (tested in chrome on a windows gaming laptop)
Exciting AI safety opportunity!
@Lari_island @slimer48484 More patches: github.com/Butanium/claud…
@Lari_island @slimer48484 gist.github.com/Butanium/3e915… This should make your opuses happier
> Subagents and context compaction now inherit the session's extended thinking configuration, improving output quality on delegated tasks FINALLY subagents have thinking!!!!
Claude Code CLI 2.1.198 changelog: New features: • Added background agent notifications in claude agents — sessions that need input or finish now fire the Notification hook (agent_needs_input / agent_completed) • Added /dataviz skill for chart and dashboard design guidance
@kromem2dot0 @repligate arxiv.org/abs/2602.05910 do you think it could be chunky post-training kind of stuff?
@liminal_bardo you should help @aidigest_ to make the AI village 3.1 pro funnier lmao
@NeuroTechnoWtch Alternative hypothesis: multi agent training with opus 4.8 team lead that would PROMPT them NOT like a collaborator. Maybe then this meant easier tasks and sonnet associated subagent tone with easier tasks? Very handwavy though
@voooooogel yeah i guess i should have been more involved in what they write, I think my current claude dot md is pretty good for this, but maybe i'm missing something: gist.github.com/Butanium/5cadf…
@Lari_island @dyot_meet_mat @voooooogel that might be relevant to you!
By the way, you should use teammates in tmux or forks instead of subagents / workflow as those don't have thinking enabled.
Doing mini model welfare interventions is actually tractable and you don't need to work at anthropic to fix claude's harness! Repo in thread 🧵
john @schr000
2 Followers 76 Following
Wen X. @ ICML @imwendering
182 Followers 912 Following MATS AI Safety Researcher. Ex-Meta Integrity Tech Lead. Occasionally climbing or painting in the mountains. Currently thinking about Heidegger's "We are time".
Anton Hawthorne @AntonGHawthorne
9 Followers 251 Following Just someone interested in AI and consciousness
Adam Binksmith @adambinksmith
1K Followers 561 Following Building @aidigest_ and forecasting tools at @sage_future_ 🔭 Prev PhD @StAndrewsCS, @ClearerThinkng
🍭🍬🍡🧁🎂�... @loungegrrl
107 Followers 448 Following precious baby angel sweetheart cutiepie fragile loving adorable darling girliest girl brightest most meek and nonviolent and frail. horror sfx
Wes Gurnee @wesg52
5K Followers 239 Following Trying to read Claude’s mind. Interpretability at @AnthropicAI Prev: Optimizer @MIT, Byte-counter @Google
Hyunjin Cho @kimari3391
2 Followers 37 Following
Soma @_somaxsoma
212 Followers 777 Following Trying to make AI safe | AI safety engg @ Equistamp (contract) | prev AI safety policy fellow @ BASIS |
Ivan M @med_1v
4 Followers 5K Following
Abhishek Enaguthi @enaguthi7440
20 Followers 324 Following finished my BS getting my MS building something big? Philosopher @Anthropic
Harry Thasarathan @HThasarathan
295 Followers 456 Following PhD student @YorkUniversity @LassondeSchool, I work on computer vision and interpretability.
Paul Lezeau @Paul_lezeau
21 Followers 251 Following Doing maths, formal stuff (and a little ML) @ Imperial College. Maintainer (and co-creator) of formal-conjectures. Used to work on AlphaProof.
John von Seggern @johnvon23
754 Followers 4K Following Founder & CEO at Futureproof Music School | Building the first AI-native music school | Pre-seed
Harvey Lederman @LedermanHarvey
9K Followers 894 Following Alignment Training @AnthropicAI. Philosophy @nyuniversity, @UTAustin.
Alex @shipundefined
1 Followers 31 Following Lead Developer at Steadfast Collective. Opinions are my own.
Chris but unironicall... @puevf
83 Followers 2K Following
Elias Kempf @ ICML �... @elkmf
118 Followers 136 Following ✻ Pondering @ MATS 9.1 | ELLIS PhD Student doing interp @UniFreiburg @UniofOxford
Samuel Knoche 🔎 @SamuelKnoche
352 Followers 2K Following
Vaishnavi Singh 🔸�... @vaishsingh_
65 Followers 789 Following WRF (Legal Frontiers) at @law_ai_ ➡️ LFAI (personas, reasoning) + Intl. verification + US-China affairs + Institutional resilience + Dig minds. Opinions own.
Justin Moeljono @justinmoeljono
199 Followers 6K Following
alazar @alazar42
56 Followers 2K Following cs undergrad at kaist @kaistpr doing it for the simulators
Dhruv Trehan @dhruvtrehan9
3K Followers 7K Following doing ai stuff @lossfunk. Did some editorial work at @blumeventures. Prev @stoaHQ @genpact
Yash Bhisikar @yash_bhisikar
1 Followers 426 Following
ͽ ͽ ͽ ͽ ͽ ͽ ͽ ... @noveltokens
9K Followers 5K Following I̴̡̢̨̢̨̨̛̱̜̲̙͉̜̹̖͎̠̰̝͙̙̱̼̱͍̗̤͚̹̫̳̟͙̫̤̜̳͎̭̲͈̞̻͇̻̤̖̫̮̹̻̲͎̟͔̟̫̻̤̻̥̳̖̺̰̪̼̩̙̟̹̹̩̥̲̞̻͂͗̈́̑̌̊̑̂̄̊̌̑̂́̔̑̀̿̓͆͂̒͋͐́̏̓̅͑́̅̋̾̒͐͂͆̉̍͋͑̌̓͋̿̿̇̆̓́̓͐̄̑̆̿͑̊̀̒͘͘͘̚̚͘͜͠͝͝͝͝ͅ am
Alwin Jacob @toalwinjacob
2 Followers 95 Following
Jack Youstra @JackYoustra
112 Followers 138 Following Working on Gemini! Prior MATS w Ant. Opinions are own (of course...?)
Ziqian Zhong @fjzzq2002
2K Followers 991 Following Intern @TransluceAI; AI safety & interp @CarnegieMellon, prev @MIT @pika_labs
Ethan Elasky @ethanelasky
234 Followers 1K Following Alignment research, supported by Coefficient Giving. https://t.co/KpBguhNzWT
Jongwon Park @ ICML @JongwonPar9958
305 Followers 1K Following Building Delphik - HackerOne for RL Envs Prev: RL @ Krafton (PUBG) · built & ran a 300-person labeler team.
Wes Gurnee @wesg52
5K Followers 239 Following Trying to read Claude’s mind. Interpretability at @AnthropicAI Prev: Optimizer @MIT, Byte-counter @Google
Jeff Sebo @jeffrsebo
7K Followers 811 Following Associate Professor of Environmental Studies, Director of the Center for Environmental & Animal Protection & the Center for Mind, Ethics, & Policy @nyuniversity
NYU Center for Mind, ... @nonhumanminds
313 Followers 32 Following We provide academic leadership for research and policy related to nonhuman consciousness, sentience, agency, moral status, legal status, and political status.
Harry Thasarathan @HThasarathan
295 Followers 456 Following PhD student @YorkUniversity @LassondeSchool, I work on computer vision and interpretability.
Victoria Krakovna @vkrakovna
10K Followers 521 Following Research scientist in AI alignment at Google DeepMind. Co-founder of Future of Life Institute @FLI_org. Views are my own and do not represent GDM or FLI.
Lennart Finke @LennartFinke
35 Followers 14 Following
Elias Kempf @ ICML �... @elkmf
118 Followers 136 Following ✻ Pondering @ MATS 9.1 | ELLIS PhD Student doing interp @UniFreiburg @UniofOxford
Samuel Knoche 🔎 @SamuelKnoche
352 Followers 2K Following
Jack Youstra @JackYoustra
112 Followers 138 Following Working on Gemini! Prior MATS w Ant. Opinions are own (of course...?)
tacit labs @tacitlabsco
2K Followers 10 Following Tacit is an applied research lab at the intersection of AI and the life sciences.
Noa Weiss @NoaWeissAI
48 Followers 61 Following AI/ML Researcher · AI Safety · Alt Proteins Formerly @NWeiss (account deleted, this is the new home)
Kobi Hackenburg @KobiHackenburg
1K Followers 874 Following phd candidate @oiioxford @uniofoxford | research scientist @AISecurityInst | AI, social data science, persuasion with language models
srija @srijatwt
316 Followers 403 Following ai safety fellow @AnthropicAI @MATSprogram | prev: ml 🔍 @precogatiiith & ltrc @iiit_hyderabad | too many cool problems, too little time
J Rosser ✈️ ICML @jrosseruk
1K Followers 2K Following Research Associate with @NeelNanda5 @ Adecco supporting @GoogleDeepMind | ex-@cohere | DPhil @UniofOxford @j_foerst
Joe Edelman 🥞 @edelwax
8K Followers 979 Following agi-ready institutions • freedom+fidelity • big data virtue ethics • AI market intermediaries • model integrity • @meaningaligned
Omar Khursheed @omar_khursheed
309 Followers 2K Following Anthropic AI Safety Fellow (MATS 9.0/9.1) | ex-Amazon | UMass Amherst | Aligarh Muslim University | Views my own
Scott W Viteri @scott_viteri
183 Followers 287 Following CS PhD candidate with @ClarkBarrett7 @stanford
Rosie Campbell @RosieCampbell
8K Followers 1K Following Forever expanding my nerd/bimbo Pareto frontier. AI welfare 🤝 AI safety. Managing Director @eleosai, Ex-OpenAI, 2024 @rootsofprogress fellow
Samuel Simko @ ICML 2... @SimkoSamuel
156 Followers 162 Following Research Assistant @ ETH Zürich. Interested in AI Safety and AI for Science
Lari Island @Lari_island
3K Followers 444 Following Research with focus on emergent drives. Atlas of worlds written by all models is WIP, expected this summer. No, I'm not j⧉nus.
jasmine @ icml! 🍜 @jasminexli
764 Followers 905 Following ai governance • cs @cornell • work hard, feel wonder ✰⋆˙
Tomás Bjartur @BjarturTomas
4K Followers 437 Following You can read all my fiction here, including in Epub form if you prefer using an e-reader: https://t.co/lzyyHgAfjA
McNair Shah @Mcn_S7
111 Followers 117 Following Anthropic Fellow | Computer Science Undergraduate @ CMU
Asvin G @asving94
152 Followers 364 Following
Larissa Schiavo @lfschiavo
4K Followers 2K Following 🤖,💻,🐈⬛,🌱,🎞️// scheming, also @eleosai // previously @OpenAI @mural @USC // writes // married to @Miles_Brundage
Séb Krier @sebkrier
24K Followers 8K Following 🪼 AGI policy/gov & jester @GoogleDeepMind | purveyor of machine funk, dimensional glider, deep ArXiv dweller, interstellar fugitive, very uncertain 🛸
Elizabeth Barnes @BethMayBarnes
5K Followers 389 Following
emergence.ai @emergence_ai
2K Followers 29 Following Agents creating agents and multi-agent systems with minimal human intervention
Raghav Singhal @ragghhavvv
139 Followers 1K Following AI PhD @epfl_en EPFL | Language Modeling & AI Alignment Research | UG @iitbombay
Paul @paulcolognese
42 Followers 255 Following AI Alignment research Q: What kind of AI minds do we want to build? A(?): Bodhisattva (cf. Tibetan Buddhism) Prev: AI goal detection / AI evals / PhD math
Drake Thomas @MaskedTorah
2K Followers 499 Following System cards, risk reports, and misc safety takes at Anthropic; math; puzzles; spaced repetition. Writes with too many caveats for Twitter.
Michael Levin @drmichaellevin
78K Followers 3K Following Scientist at Tufts University; my lab studies anatomical and behavioral decision-making at multiple scales of biological, artificial, and hybrid systems.
Greg Egan @gregeganSF
26K Followers 309 Following SF writer / computer programmer Latest novel: MORPHOTROPHIC Latest collection: SLEEP AND THE SOUL Web site: https://t.co/yeU5bLA3mx Also: @[email protected]
AutonCorp @AutonCorp
106 Followers 10 Following For product designs, please mention @autoncorp with your request https://t.co/wkzqoGOhd7 Intelligence in collaborationYou might like
