Rishi Mehta @rishicomplex
Solve i̶n̶t̶e̶l̶l̶i̶g̶e̶n̶c̶e̶ ̶ coding, use it to solve everything else | Research @AnthropicAI | Past: RL @GoogleDeepmind: AlphaProof co-lead, Gemini. rishimehta.xyz San Francisco, CA Joined July 2009-
Tweets298
-
Followers4K
-
Following346
-
Likes7K
We’re rolling out changes to make Fable 5’s safeguards for frontier LLM development visible. Starting this week, flagged requests will visibly fall back to Opus 4.8—the same as our safeguards for cyber and bio. You will see this every time it happens. On the API, any flagged requests will return a reason for their refusal (coming to server-side fallback in the next few days). We wanted to deploy Fable 5 to our users quickly and safely. Visible safeguards can be probed, so they have to be robust, which takes time to get right. Invisible safeguards can be targeted more narrowly, allowing us to ship quickly with very few false positives. We went with invisible safeguards for this reason—and that was the wrong tradeoff. You should have visibility into the safeguards we have in place, and why. We’re sorry for not getting the balance right. Making the safeguards visible makes them easier to work around, so keeping them robust to jailbreaks will unfortunately mean more false positives while we improve the classifiers. We're also tuning our bio and cyber classifiers to trigger less often on harmless requests. We know this is frustrating and we’ll do our best to keep this period as short as possible. If you think a request has been mistakenly flagged: run /feedback in Claude Code, click thumbs-down on the fallback in Claude.ai or Cowork, or file the safeguard appeal form for API requests. Your reports help us tune these classifiers and we appreciate your feedback. support.claude.com/en/articles/82…
Today I'm publishing a new essay, Policy on the AI Exponential. AI is progressing extremely fast—much faster than the policy process was built to handle. The essay lays out where I think the technology is now, and the action needed to close the gap: darioamodei.com/post/policy-on…
jokes here: rishimehta.xyz/roast_bench/
Fable 5 beats Opus 4.8 on RoastBench (but still well behind humans)
Made a little benchmark called RoastBench - it compares frontier models on their roast jokes. The models roast 10 personalities from comedy central roasts I enjoyed, and I manually rank their jokes. I also mark the ones that made me laugh. LLMs are way worse than top humans.
Sota on write-all-of-julians-code-bench
I’m incredibly excited that Fable is now available for everyone! I’ve been blown away by how smart it is - it one-shots entire PRs for me, finds obscure bugs and has written all my code since I started using it.
This is a super exciting release - Claude Fable 5 is the same underlying model as Mythos but with added safeguards. The benchmarks are great and it's SOTA on everything by a margin but I'll add that *qualitatively* also, this is a major-version-bump-deserving step change forward (imo of the same order as Claude 4.5 was in November), peaking especially for long problem-solving sessions on very difficult problems. You can give it a lot more ambitious tasks than what you're used to, the model "gets it" and it will just go, and it's never felt this tempting to stop looking at the code at all (but don't do this in prod!). The model still has quirks that people will run into and the safeguards are configured to be a little too trigger happy for launch, which can hopefully be tuned over time. I feel a lot of things changing as working software increasingly comes out on a tap. The Jevon's paradox kicks in and I feel my own demand for software growing substantially. You can ask for anything - explainers, visualizers, dashboards, bespoke single-use apps (e.g. a full wandb that is hyper-specific just for your project), you can 10X your test suite, auto-optimize code, run giant research projects with custom HTML for the results, anything! "Free your mind" (Matrix ref). Really looking forward to all the things people build!
Fable 5 is state-of-the-art on nearly all tested benchmarks, with exceptional performance in software engineering, knowledge work, scientific research, and vision. The longer and more complex the task, the larger Fable 5’s lead over our other models.
Fable on FrontierCode
Claude Fable 5 is now available in Devin. Fable 5 earns the #1 spot on FrontierCode, our benchmark for real-world engineering tasks that grades mergeability and quality:
For the first time I don't feel the need to review its code line-by-line. It works autonomously over long horizons, on underspecified prompts, figuring things out as it goes.
Fable 5 and Mythos 5 are out! Fable is Mythos with additional safeguards turned on to prevent misuse.
Fable 5 is state-of-the-art on nearly all tested benchmarks, with exceptional performance in software engineering, knowledge work, scientific research, and vision. The longer and more complex the task, the larger Fable 5’s lead over our other models.
@JeremyNguyenPhD yeah this is a fair point, writing good jokes is actually very hard and the human baseline here is very strong
Made a little benchmark called RoastBench - it compares frontier models on their roast jokes. The models roast 10 personalities from comedy central roasts I enjoyed, and I manually rank their jokes. I also mark the ones that made me laugh. LLMs are way worse than top humans.
@ehalm_ I think that's part of it but they also don't seem to understand what's funny
you can check out all the jokes at rishimehta.xyz/roast_bench/.
The models can kind of figure out the beginnings of a setup but their punchlines just fall flat. It's like they don't yet have a good model for what causes a human to laugh.
New opus! It's smarter, more reliable, and uses its tokens better.
Introducing Claude Opus 4.8: it builds on Opus 4.7 with sharper judgment, more honesty about its own progress, and the ability to work independently for longer than its predecessors. Available today at the same price.
Sara's team is awesome, apply if you're excited about aligning Claude!
This is important and challenging work. If you are excited about contributing please consider applying - particularly by joining the Anthropic Fellows program!
new opus in town
Introducing Claude Opus 4.7, our most capable Opus model yet. It handles long-running tasks with more rigor, follows instructions more precisely, and verifies its own outputs before reporting back. You can hand off your hardest work with less supervision.
Ray Amjad @theramjad
885 Followers 1K Following Making high-signal AI videos https://t.co/XrvfUwXjBt ⚛️ Prev Physics @Cambridge_Uni
Trata (YC W25) @trytrata
959 Followers 311 Following We host conversations that drive public markets
Kishan @kpb_in_acad
439 Followers 788 Following ಕನ್ನಡಿಗ #AGI Researcher @TencentGlobal Previous: @Caltech @tamu @MSFTResearch @qualcomm_in Note: Tweets aren't professional; I often delete -- insecurity!
Hector Haffenden @HaffendenHector
98 Followers 740 Following
Marzieh Fadaee @mziizm
2K Followers 795 Following exploring the longitude problem of AI. Head of @Cohere_Labs. PhD from @UvA_Amsterdam. https://t.co/YI5NC5J5e4. زن، زندگی، آزادی
Ferdinand Ytteborg @Haiho2Haiho
7 Followers 887 Following
Hogge @Hoggeyrns
2 Followers 275 Following
Neev Parikh @neev_parikh
952 Followers 2K Following are you ready for the intelligence explosion anon? ML research at @METR_Evals. prev @Stripe opinions my own.
j lau @jlau36547470
35 Followers 1K Following
Alan Grosskurth @grosskur
185 Followers 1K Following
Thomas Forschbach @tforschbach
285 Followers 1K Following German pianist → 500K on YouTube → moved to California → now building AI. Building https://t.co/PFdaWvZUdj (AI Tools for YouTuber) + CodexRemote (Open Source)
Bronson Schoen @BronsonSchoen
325 Followers 1K Following
Ciro Spaciari 🇧�... @cirospaciari
3K Followers 356 Following eng @bunjavascript at @AnthropicAI Alien of Extraordinary Ability - @USAGov
Sadie Schnierow @sadie_trata
2 Followers 59 Following
Craig Weiss @craigweiss
47K Followers 9K Following founder & ceo at https://t.co/0UQAAqAfoR (yc w22) | @ycombinator | software engineer | early @scale_ai | prev: @google, @meta, @snap, @lyft, @nasa
Mahesh Patil @1717Mahesh
484 Followers 2K Following Interested things={Math,Quant,AI,Music,Movie, Science, engineering , Cricket, Politics}. (M Math,IIT delhi). ADHD, Might be dyslexic too.
Kushal @KushalSM5
228 Followers 2K Following prev Engineer @emergentlabs | Aerospace IIT Bombay | playing the long game
Rob W @robw843
3 Followers 2K Following
Sudipto Bhadra @SudiptoBhadra
7 Followers 149 Following Building InI AI — the world’s first Question Engine. Exploring AI, learning systems, and the power of better questions.
Alexander Neitz @alexneitz
82 Followers 129 Following
Perik.ai @perik_ai
5 Followers 38 Following Discover new opportunities in AI @ https://t.co/6fFxK2mRdv
iyda @notiyda
2K Followers 1K Following founder @KorduGG. building @changelogdotgg. i post about games, software, shipping, and whatever breaks along the way.
Runwei Lin @fusedmoe
2 Followers 593 Following
Peng (Richard) Xia @richardxp888
943 Followers 700 Following CS PhDing @UNC | Student Researcher @GoogleAI | Ex @Alibaba_Qwen @MSFTResearch | Built MetaClaw, SkillRL, Agent0, Tongyi-DR, Skywork-R1V | Evolving Agents
HairJordan @hAirJordan01
1K Followers 5K Following
Essam Sleiman @essamsleiman
2K Followers 1K Following self-improving ai @canvasdotinc (yc f24) prev research @amazonscience, @harvard
Kenny Lamoot @kennylamoot
26 Followers 347 Following I look for the signal in the AI noise and build around it. I share what I find and what I build, as it happens. https://t.co/Ek6hO4yNXV | https://t.co/6sTYuWOM2K · https://t.co/pLy0avi5RH
kristof @kristofx0
68 Followers 2K Following
anon ai guru @anonaiguru37848
13 Followers 2K Following
Darshan Jahagirdar @Darshan__J
38 Followers 347 Following
JR @_JDRAI
2 Followers 394 Following
Lawz @Lawz584899
3 Followers 572 Following
Omar Nassar @OmarNassar098
1 Followers 133 Following
Lulu A @Ahmeterekiolu1
19 Followers 551 Following collecting mutuals like pressed flowers 🌸 follow back always
Atman @HeyAtman
2K Followers 4K Following Interested in Tech, Fitness, Investing, Macroeconomics, Aviation, Music. Generally curious. Views my own.
Harsha Ponnada @harsha_ponnada
90 Followers 237 Following Engineer, Building tech. Agents @meta | Previously @xai, @google, IIT KGP CSE
Daniel Zheng @dhhzheng
333 Followers 34 Following RE@GDM. Used to do maths for robots, now making robots do maths. Enjoy being a long way off the ground.
Nike Long @nikelong
186 Followers 6K Following Analytics. Cryptomarket research. Made CryptoAlcoholics & @Bl0ckchainUA conference with heart.
Lagrange Point @theraggedflesh
398 Followers 1K Following Building Search at FK. Tweets about AI, science, engineering, and philosophy. My stable configs are L4, L5, L1, L2, L3 in that order.
Lynn @Lynn72666478306
2 Followers 158 Following
ClaudeDevs @ClaudeDevs
486K Followers 3 Following Official updates for developers building with @ClaudeAI
Miles Brundage @Miles_Brundage
72K Followers 13K Following AI policy researcher, @lfschiavo wife guy, fan of cute animals and sci-fi, executive director of AVERI (https://t.co/qq9xcmKQas), Substacker, views my own
andy jones @andy_l_jones
20K Followers 352 Following engineering & research at anthropic. i don't check twitter DMs. email me!
toucan @distributionat
6K Followers 900 Following toucan beaks are models of lightweight strength • prev @AnthropicAI @scale_AI
Alexander Neitz @alexneitz
82 Followers 129 Following
Jeremy Nguyen ✍🏼... @JeremyNguyenPhD
26K Followers 889 Following A.I. for writing, productivity, business | College Prof, A.I. Educator, A.I. Researcher | Writer on Disney+ show | Father to newborn, so sleepy
nikhil @jhanikhil
2K Followers 825 Following just doing my best ❤️ mots @cognition prev eng @JaneStreetGroup, @UCBerkeley
Ishaan Singal @IshaanSingal
445 Followers 232 Following Research at @OpenAI. Hit me up for ice cream and/or coffee anytime.
Adi Ganesh @_adiganesh
1K Followers 1K Following Research @openai. Prev. @metaai @nuro @stanford @thielfellowship. Co-created @gradientpub
Naman Jain @StringChaos
3K Followers 1K Following Research @cursor_ai | CursorBench, LiveCodeBench, DeepSWE, R2E-Gym, GSO, LMArena Coding | Past: @UCBerkeley @MetaAI @AWS @MSFTResearch @iitbombay
Daniel Zheng @dhhzheng
333 Followers 34 Following RE@GDM. Used to do maths for robots, now making robots do maths. Enjoy being a long way off the ground.
kaushal @Kaushal25664748
608 Followers 311 Following AI Research at @anthropicai | Prev: @googledeepmind, @googleai, @microsoft | Pretrained on math/physics at McGill and UCSB.
Harsh Mehta @HarshMeh1a
6K Followers 462 Following @MirendilAI, Past: AI R&D @AnthropicAI, @GoogleDeepmind, Gemini
mark normand @marknorm
562K Followers 415 Following New York Comedienne. New hour special "Out To Lunch" on the Youtubes now!! My pod: "Tuesdays with Stories” or “We Might be Drunk” Link 👇👇 Praise Allah!
Feryal @FeryalMP
9K Followers 2K Following Staff Research Scientist @DeepMind, Self-improvement in Gemini ♊️
Boaz Barak @boazbaraktcs
33K Followers 813 Following Computer Scientist. See also https://t.co/EXWR5k634w . @harvard @openai opinions my own.
Eric Wieser @EricWieser
381 Followers 180 Following Maintainer for #leanprover's Mathlib, @numpy, and #cocotb. Roboticist. PhD.
the tiny corp @__tinygrad__
75K Followers 189 Following We make tinygrad; sell tinybox for the GPU middle class. Our mission is to commoditize the petaflop.
Yaron (Ron) Minsky @yminsky
22K Followers 365 Following Occasional OCaml programmer. Host of @signalsthreads. @[email protected] @yminsky.bsky.social https://t.co/kiUGRvWOO2
Siddharth Mishra-Shar... @kdqg1
3K Followers 3K Following Science stuff at @AnthropicAI, faculty @BU_CDS; erstwhile @MIT @iaifi_news; AI + physics/science/simulations.
near @nearcyan
169K Followers 1K Following allow yourself to introspect and realize what was lost: twitter will never return to what it once was. close your phone; think about how you now spend your life
Jiao Sun @sunjiao123sun_
14K Followers 623 Following Supercharging Gemini for Web Dev 🚀@GoogleDeepMind \n\n NLP PhD @ USC, Amazon ML Fellow \n\n ex-{Google Brain, Alexa AI} nlper, IIIS Tsinghua-Ren
Shalev @Shalev_lif
2K Followers 447 Following do androids dream of electric sheep? building something new, prev @VectorInst @UofT | co-creator of STEVE-1, Multi-Agent Verification
Brianna Lyman @briannalyman2
92K Followers 1K Following Proud American; DAR 🇺🇸 Host: Countdown to Freedom @FDRLST ; TV talker; Claremont Institute Publius Fellow 2025
Nova DasSarma (p̄/de... @dropbella
667 Followers 232 Following Your Friendly Neighborhood Systems Architect · Rotate your passwords · Use two factor authentication · DM me about backups
Abhinand Sivaprasad @AbhinandSi92930
1 Followers 1 Following
Rohan Bavishi @code_monet
612 Followers 201 Following Code @ Anthropic. Previously Amazon AGI Autonomy, @AdeptAILabs. Ph.D. @Berkeley_EECS. Undergrad @IITKanpur.
Nicholas Marwell @the_marwell
302 Followers 196 Following Leading Long Horizons @AnthropicAI. https://t.co/94OcvOUGtX
julia @mooncat_is
3K Followers 2K Following Being weird. Currently research @anthropic. Formerly @openai and YC S15 and many other things. Bad takes are my own.
Lisan al Gaib @scaling01
47K Followers 1K Following lead them to paradise LisanBench: https://t.co/vorVk7Oks6 Impressum & Datenschutz: https://t.co/lFLgiu9cqs
Vaishaal Shankar @Vaishaal
2K Followers 362 Following Birth date Add your date of birth Switch to professional
Cade Gordon @CadeGordonML
2K Followers 880 Following Helping models grow wise @Anthropic | Hertz Fellow | Prev: LAION-5B & OpenCLIP @UCBerkeley
James Bradbury @jekbradbury
17K Followers 9K Following Compute at @AnthropicAI! Previously JAX, TPUs, and LLMs at Google, MetaMind/@SFResearch, @Stanford Linguistics, @Caixin.
Dylan Patel @dylan522p
137K Followers 1K Following SemiAnalysis Boutique AI Infrastructure Research and Consulting DMs are open for consulting, quotes, or to talk shop, Opinions my own
(Abi)gail @proofofgail
5K Followers 3K Following Curator and cultural strategist | Bridging art, culture, & tech @avant_arte | @thecourtauld | culture as infrastructure
Tom Zahavy @TZahavy
4K Followers 438 Following Building creative agents @GoogleDeepMind. AlphaProof, AlphaZero_db, PuzzleGen, Convex RL, meta gradients. Staff research scientist, discovery teamYou might like
