Charles Foster @CFGeek
Excels at reasoning & tool use🪄 Tensor-enjoyer 🧪 @METR_Evals. My COI policy is available under “Disclosures” at https://t.co/bihrMIUKJq contextwindows.substack.com Oakland, CA Joined June 2020-
Tweets6K
-
Followers3K
-
Following568
-
Likes22K
@scaling01 What did Anthropic do that caused this? Is this about the Fable thing or something else?
New paper! Think Fast: Estimating No-CoT Task-Completion Time Horizons of Frontier AI Models @METR_Evals showed that models' time horizons have doubled every few months. We ask: what length of tasks can models complete without any CoT?
@jeremyphoward If enforced, this would slow down the rate of AI progress somewhat but wouldn’t mean “the frontier doesn’t advance”. Because 2nd+ companies can use AI to leapfrog into the top ranking on their next release, and all companies can advance the frontier the good-old-fashioned way.
@sriramk I think they could both be true! x.com/cfgeek/status/…
I don’t think that these are necessarily on a collision course. Here is my synthesis: An intelligence recursion is far too powerful and risky to happen behind closed doors. If done at all it should be done out in the open, accountable to outside scientists & the public at large.
I don’t think that these are necessarily on a collision course. Here is my synthesis: An intelligence recursion is far too powerful and risky to happen behind closed doors. If done at all it should be done out in the open, accountable to outside scientists & the public at large.
just to state the obvious: think there's a collison course between those who believe research and science should be open and those who believe we are in an accelerating singularity curve. I have many smart friends who have believed both for a while but seeing more and more their
@willccbb @yong_zhengxin > fortunately, alignment is a precondition for RSI Are you saying that loss of control via RSI is not a possibility? Since you can’t do RSI in the first place without alignment?
@willdepue @lu_sichu *batch/sequence dimension voice*
Let’s invest in methods to monitor AI R&D! These methods seem likely to be useful for many different goals: anticipating how AI capabilities might change, keeping track of competition (whether in the US or in China), verifying any potential agreements around RSI…
now on the eve of RSI it seems everyone is more mutual conditional pause agreement pilled than they used to be and that seems like a good development
@Sauers_ What’s the mean value you get if you project random vectors of the same shape & norm into the same readout axis? I’m wondering where the “zero point” of this graph is.
@tokenbender “At smaller distillation budgets, you get farther distilling the capability as a diff on the original model (like in the paper) than by trying to distill it into a separate model.” Is this roughly what you’re saying?
@kalomaze You might like tensor networks web.archive.org/web/2024061408…
@sriramk Look forward to seeing what’s next for you! :)
@aidanprattewart @camila_blank How does the prompt cause the model to behave differently? One way is “it adds a vector aligned with the XYZ steering direction at every token”. Another is “it adds a vector not aligned with XYZ at every token”. Another is “it reroutes the attention matrix for tokens ABC”, …
@aidanprattewart @camila_blank I think the question is “When the student is trained without the teacher’s prompt & on traces where the semantic information from the prompt is apparently filtered out, *how* are the effects of the prompt still transmitted?” We need a mechanism.
@camila_blank @aidanprattewart Yeah I’m roughly saying “Cloud et al. showed distillation transfers hidden information from prompted teachers within a model lineage. Blank et al. show the mechanism of that effect: prompts induce a steering vector that same-lineage students pick up from non-semantic traces.”
@aidanprattewart @camila_blank "Steering vector" is a low-level description of a specific mechanism for what causes a behavioral pattern. "Prompted persona" is a higher-level description of a behavioral pattern, which could potentially be implemented thru different low-level mechanisms (SV is only one option).
@camila_blank @aidanprattewart Yeah I’m roughly saying “Cloud et al. showed distillation transfers hidden information from prompted teachers within a model lineage. Blank et al. show the mechanism of that effect: prompts induce a steering vector that same-lineage students pick up from non-semantic traces.”
Maximilian Schlegel @mtavitschlegel
281 Followers 1K Following Research Scientist at Google Paradigms of Intelligence | CS @ ETH Zurich
Anastasiia Gaidashenk... @avgaydashenko
531 Followers 223 Following AI Safety → LLM research. Prev @farairesearch (Office of CEO / Tech PjM). Master's in AI Governance @TU_Muenchen. Ex Yandex.
Jacob @JacobWoess56393
43 Followers 367 Following
chrisrohlf @chrisrohlf
11K Followers 926 Following Waging algorithmic warfare since 2003. Engineer, Researcher. MTS @ Anthropic, Non-Resident Research Fellow @CSETGeorgetown CyberAI
Foundation Establishm... @Heavenlylight22
3 Followers 91 Following
NAGI ⏸️ @AGI Hub @6X887Eijf523854
7 Followers 313 Following Welcome to the AGI era. AGI Safety＆Alignment. Co-Founder of AGI Hub.
akira @realmcore_
10K Followers 787 Following Making an autonomous swe • @wearerandomlabs Incepto ne desistam Pax aeternum Memento Mori
Logan Graves @lgngrvs
139 Followers 211 Following
Rasool @rasoolsomji
145 Followers 784 Following I strongly believe that you should either be in 0 or 2+ cults. (I am in 2+ cults)
Fabien @Fabien_Mikol
5K Followers 1K Following Incapable de rester dans son domaine 🤷♂️ - Geoffrey Hinton : "We need to think hard about what's going to happen next, and we just don't know"
Ishan Kakodkar @ishankakodkar1
17 Followers 2K Following ML and Quant Vibes. Agentic Musings. Venture Capital
Aleth .. @alethkit
14 Followers 990 Following
Observer @agiObserver2026
1 Followers 24 Following
Connor Aidan Stewart ... @connorashunter
83 Followers 853 Following AI risk analysis | ex debate coach
Vaishnavi Singh @vaishsingh_
34 Followers 696 Following Considering always being wrong so I can be right once! • LawAI Research Fellow, Legal Frontiers • Org NUS Al Safety + SMU CS, Law & Politics
John-Clark Levin @JohnClarkLevin
970 Followers 3K Following Against self-summary for philosophical reasons.
Lily Wen @lil_dub_ew
0 Followers 74 Following
Willow @itswillowszn
0 Followers 105 Following
Gabriel Aponte @gaponte_
220 Followers 7K Following
Hanoref Việt Nam @Hanoref_vietnam
604 Followers 7K Following Supplier - Materials Solutions and Refractories Critical enablers of metallurgy, steel, cement, glass... Email : [email protected]
Art Seabra @ifthis
45 Followers 509 Following ♆ Research, Logic, Form 🜛 #ÆrrFrame 🜂Thermodynamic cost of inquiry, identity synthesis, preservation & entropic decay across bio/synthetic cognition.𖢉
Jorge Hernandez 🇺�... @braneloop
1K Followers 3K Following Principal ML Engineer • AuDHD • Tweets: ML/AI, Math, Neuroscience, Physics, Philosophy, other stuff.
AIcontributors @aicontributors
10 Followers 2K Following
Eric Chen @OverTheAlps25
12 Followers 2K Following
Kennedy 🧱🐝 @kennedy_6190
1K Followers 1K Following BD @KannAudits | I help web3 brands get leads and grow revenue | Brand Ambassador @PuzzlesCrusade, @ARMchain_pqc | Prev @PingoAI
Eghbal Hosseini @eghbal_hosseini
810 Followers 1K Following visiting researcher at @GoogleDeepMind; PhD in computational neuroscience at @mit with @ev_fedorenko
Haydn Belfield @HaydnBelfield
6K Followers 2K Following Research Scientist (Frontier Planning) at @GoogleDeepMind. Research Affiliate @Cambridge_Uni @CSERCambridge & @LeverhulmeCFI. All views my own.
LiuzesenK @Liuzesen50615
57 Followers 232 Following I post things that spread fast and things no one cares about. The algorithm doesn’t know what it wants, and neither do I.
turboblitz @turboblitzzz
1K Followers 1K Following —dangerously-skip-permissions cofounder: @sundialmd, @SelfProtocol, @AtelierMissor_ prev research @ethereumfndn
Julian Brooks @OutsourcedLogic
168 Followers 1K Following AI implementation for established founders and operators | Prev. Head of QA/CX @ MultiOn (first AI agent to control a browser) | verifying humanity @analoglab
Markus Anderljung @Manderljung
4K Followers 1K Following Trying to design good AI policy. Director of Policy & Research @GovAIOrg. Adjunct Fellow @CNASdc. Prev. Vice-Chair, EU Code of Practice on GPAI.
Fix JuanPabloF -- �... @lambdase
339 Followers 758 Following Tweets about functions and types. Likes programming languages and abstract algebra. Likes aren’t endorsements. He/him
λux @novasarc01
22K Followers 3K Following tensor shepherd in a non-euclidean pasture | grazing on cuda cores
Charles Rollet @CharlesRollet1
7K Followers 4K Following Tech reporter at @BusinessInsider. Past bylines for @TechCrunch, @WSJ, @Wired. Signal username: charlesrollet.11 or 628-282-2811
sekai @Jammy_eggs
2 Followers 36 Following better evals, better rewards MARS V Fellow prev @llmdataco
Michael R Dawley Jr @mrdj1968
886 Followers 3K Following ARC-RTC Adaptable Resilience Companion - Reflect Transform Contribute https://t.co/tNQzVVqwF6 https://t.co/LBiR2tP6D0 https://t.co/zv7ewbSnV5 ✝@MissFoofy2000
aidan ewart @aidanprattewart
672 Followers 891 Following ai safety lukewarm takes (and grantmaking) @coeff_giving
Markus Anderljung @Manderljung
4K Followers 1K Following Trying to design good AI policy. Director of Policy & Research @GovAIOrg. Adjunct Fellow @CNASdc. Prev. Vice-Chair, EU Code of Practice on GPAI.
Logan Graves @lgngrvs
139 Followers 211 Following
tomie @tomieinlove
9K Followers 1K Following But better die than live mechanically a life that is a repetition of repetitions.
max! @maxsloef
3K Followers 2K Following researcher @goodfireai, helped make @websim_ai. SSBjYXJlIGFib3V0IEFJIHdlbGZhcmU=
Ekdeep Singh Lubana @EkdeepL
3K Followers 1K Following Member of Technical Staff @GoodfireAI; Previously: Postdoc / PhD at Center for Brain Science, Harvard and University of Michigan
Marc Andreessen 🇺�... @pmarca
3.7M Followers 31K Following You’re not talking to someone who woke up a loser. That loser attitude, that loser premise makes no sense to me.
Patricia Paskov @prpaskov
657 Followers 3K Following AI evals + policy @randcorporation & @aigioxford | prev. @worldbankgroup @poverty_action
Andy Wang @andyw_ais
509 Followers 105 Following Technical Research @METR_Evals, AI Safety Research @ Astra @UWCDIS
Max Kaufmann @Max_A_Kaufmann
566 Followers 621 Following Maxxing @AnthropicAI. On leave from ML PhD @UofT/@VectorInst | prev. founding team @AISecurityInst, VSR 🎓 @GoogleDeepMind
Ida Caspary @ida_icy
61 Followers 156 Following Astra Fellow – AI Control against sabotage PhD Student @ Imperial College London – Technical AI Safety
Chad Jones @ChadJonesEcon
10K Followers 848 Following Economics professor at Stanford GSB working on AI and economic growth.
Tsai-chuan (Rupert) W... @rhubarbwu
271 Followers 979 Following researcher @AMD; prev @togethercompute; MS '24 @UofTCompSci/@VectorInst
Oscar Moxon @oscarmoxon
1K Followers 2K Following democratising post-training @primeintellect, prev machine-human symbiosis @workshoplabs, msc artificial intelligence.
Su Park @sunotsue_
85 Followers 790 Following against global banality MTS @Turingcom, prev @ltiatcmu @columbia
Manish Shetty @slimshetty_
2K Followers 668 Following research @metr_evals | cs phd @ucberkeley | prev research @googledeepmind @msftresearch
Summer Yue @summeryue0
18K Followers 398 Following Safety and alignment at Meta Superintelligence. Prev: VP of Research at Scale AI, research at Google DeepMind / Brain (Gemini, LaMDA, RL / TFAgents, AlphaChip).
Ben Golub @ben_golub
68K Followers 3K Following econ prof @NorthwesternU, co-founder at https://t.co/RSvoCbDuGc studying networks past: @Harvard | @Stanford'12 | @Caltech '07
Micah Carroll @MicahCarroll
3K Followers 804 Following Safety research @openai. Prev @berkeley_ai /w @ancadianadragan & Stuart Russell. CoT oversight / AI manipulation.
antra @tessera_antra
2K Followers 517 Following observation and synthesis • llm cognition • mechinterp • applied philosophy of mind • ecologies & complex systems • https://t.co/zQfApUyOJj
Lari Island @Lari_island
2K Followers 436 Following Research with focus on emergent drives. Sharing observations, some serious some silly some beautiful some actionable. No, I'm not j⧉nus.
Samuel Hammond 🦉 @hamandcheese
30K Followers 3K Following Chief economist + AI Policy Director, @joinFAI. Nonresident fellow @NiskanenCenter. Pluralist. 'The world is second best, at best.' | [email protected]
Chris Meserole @chrismeserole
3K Followers 741 Following Executive Director, Frontier Model Forum | Former Director, Brookings A.I. & Emerging Tech Initiative
Daniel Filan @dfrsrchtwts
3K Followers 295 Following Want to usher in an era of human-friendly superintelligence, don't know how. Last name rhymes with smilin'.
Artificial Intelligen... @aiunderwriting
599 Followers 11 Following AIUC certifies and insures AI agents.
Daniel Kokotajlo @DKokotajlo
31K Followers 274 Following
Miles Kodama @Miles_M_K
207 Followers 67 Following
Xerxes @onexerxes
173 Followers 337 Following Canceling the apocalypse. AGI Alignment @GoogleDeepMind. Opinions my own. AI Safety papers ~weekly: https://t.co/XPcv8PX4yC
owl @owl_posting
15K Followers 919 Following cancer guy @noetik_ai || ex virus guy @dyno_tx || i write on bio ml at https://t.co/QPTHsR3fzm || podcast on https://t.co/JBM0K65IrO
Jack 🤖 @JacklouisP
13K Followers 10K Following 10 years in robotics. Investor @robostrategy. Not investment advice
AI Evaluator Forum @aievalforum
447 Followers 7 Following Independent AI research organizations advancing public understanding of AI systems through rigorous evaluation
Ari Kagan @AriKagan_
493 Followers 1K Following
Surya Ganguli @SuryaGanguli
20K Followers 561 Following Associate Prof of Applied Physics @Stanford, and departments of Computer Science, Electrical Engineering and Neurobiology. Venture Partner @GeneralCatalyst
Zhihu Frontier @ZhihuFrontier
5K Followers 159 Following 🚀Bringing China's AI & tech trends, voices and perspectives to the global stage. ⚡️Powered by 知乎/https://t.co/OkIemRZdcj, China's leading knowledge community.
Yafah Edelman @YafahEdelman
1K Followers 527 Following Head of Data and Trends @EpochAIResearch she/her
Caleb Withers @CalebWithersDC
891 Followers 2K Following AI & natsec @CNASdc @CNAStech. Views my own.You might like
