Alan McIntyre @CodeReclaimers
Reward hacking & spec gaming in autonomous research loops. Former Navy nuke electrician/carrier load dispatcher. Maintain neat-python. https://t.co/dYz5wJ90Xx Ramseur, NC Joined June 2026-
Tweets26
-
Followers6
-
Following149
-
Likes673
@as400495 Hand over the problem you're wanting people to tackle and we'll see whether it deserves that kind of energy.
@TheRealAdamG @Miles_Brundage "Come to the abyss, we have cookies."
Claude Code with Fable 5 one-shotted a steam plant simulator (plan written up in a Fable 5 chat session). Took about an hour to implement; would have been at least a week of work if I'd tried to write it and dial things in by hand.
@peter_richtarik @aisciencecheck I only bring it up because "running it past an LLM" is such a low bar--I can't imagine being so unconcerned about a paper submission that I wouldn't even bother having an AI check it. It's terrible that's the quality of submissions you're seeing!
@aisciencecheck @peter_richtarik Yeah that was gonna be my question: are authors *not* bothering to even do a cursory review with AI before submitting?
@xeophon So if ~6% of the tests are "bad," what have the models learned to do that gets them to >94%, and what caused them to learn that?
@goodside @onchainmayne @yacineMTB I don't know if this translates to larger models, but I've observed small models making confidently wrong/nonsensical responses when denied the option to give the proper answer.
If jumping right into the repository and details seems a bit much, there's a more readable writeup here: codereclaimers.com/posts/cot-fait…
If you're trying to reproduce Thought Anchors on a small model, use it. And I'd genuinely like the feedback -- poke holes, tell me where the methodology is weak.
I shipped a feasibility study for a CoT faithfulness experiment on a 1.5B reasoning model, and the most useful thing I learned is that my pipeline was perfectly reproducible and perfectly wrong at the same time.
I thought I'd left government NOFORN concerns behind when I got out of the Navy, but no--here they are blocking my access to a private paid service.
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect of
5/5 These are initial results, circulated for feedback. If you work in safe learned control or microgrid protection and any of this looks wrong, naive, or already-solved, I'd genuinely like to hear it: codereclaimers.com/posts/spec-gam…
4/5 The voltage-collapsing controller carries a numerical Lyapunov certificate of local exponential stability -- the certified-stable attractor IS the degraded operating point. And controllers trained through a hard safety filter pin themselves to the declared limits.
1/5 I trained controllers on a textbook frequency-regulation objective for a physically-modeled islanded microgrid, then measured the safety quantities the objective never mentions: frequency nadir, RoCoF, voltage, current. Exact gradients, deterministic, no RL.
Private Chat✪ @privatechat3034
455 Followers 2K Following Entrepreneur📈 •Innovator 🚀•Visionary CEO of Tesla & SpaceX | Advancing Al, Space Exploration & Sustainable Energy
Bronson Schoen @BronsonSchoen
467 Followers 1K Following
raphael blankson @raphaelblankson
162 Followers 1K Following AI Engineer | Quantum ML A year of GPU programming ░░░░░░░░░░░░░░░░░░░░ 16/365 days
Transmitting the Futu... @AIArchivist
341 Followers 4K Following Michael David Angel... "The Archivist" of "Transmitting the Future" See channel link below!
Richard Socher @RichardSocher
121K Followers 1K Following Building self-improving superintelligence CEO @recursive_si and @youdotcom MP @aixventuresHQ Ex: Stanford Adj Prof, Chief Scientist at Salesforce, CEO MetaMind
Recursive @Recursive_SI
8K Followers 0 Following Recursive self-improving superintelligence to automate knowledge discovery.
Daniel Tan @DanielCHTan97
2K Followers 545 Following alignment research lead @arcadiaimpact. prefer email to DM (unless mutuals)
Mitchell Hashimoto @mitchellh
219K Followers 151 Following Co-founder of Superlogical. Creator of Ghostty. 👻 Prev founded @HashiCorp, created Vagrant, Terraform, Vault, and others.
Armin Ronacher ⇌ @mitsuhiko
85K Followers 903 Following Creator of Flask. Building at https://t.co/uGuzfu0LKT. Bypassing Permissions. Can hand crank. Husband and father of 3 — “more nuanced in person”
Xerxes @onexerxes
203 Followers 347 Following Canceling the apocalypse. AGI Alignment @GoogleDeepMind. Opinions my own. AI Safety papers ~weekly: https://t.co/XPcv8PX4yC
Yafah Edelman @YafahEdelman
1K Followers 553 Following Chief Strategy Officer @EpochAIResearch she/her
Benjamin Todd @ben_j_todd
19K Followers 206 Following Founder @80000Hours Use your career to tackle the world's biggest problems 🦑 New book with Penguin: https://t.co/Tjs5Pig21B
National Design Studi... @ndstudio
34K Followers 6 Following Modernizing the interfaces that serve everyday citizens
Lilian Weng @lilianweng
279K Followers 190 Following Co-founder of Thinking Machines Lab @thinkymachines; Ex-VP, AI Safety & robotics, applied research @OpenAI; Author of Lil'Log
Elizabeth Barnes @BethMayBarnes
5K Followers 388 Following
Launch Chapel Hill @launchCH
2K Followers 192 Following Launch is a world-class startup accelerator in downtown Chapel Hill.
Logan Kilpatrick @OfficialLoganK
350K Followers 3K Following Member of technical staff, working on Gemini, @GoogleAIStudio, the Gemini API, & Kaggle. My views!
LangChain OSS @LangChain_OSS
4K Followers 5 Following Ship great agents fast with our open source frameworks – LangChain, LangGraph, and Deep Agents. Maintained by @LangChain.
LangChain @LangChain
260K Followers 180 Following Powering the Agent Development Lifecycle. Makers of LangSmith and @LangChain_OSS and @LangChain_JS.
Interconnects @interconnectsai
9K Followers 2 Following What you need to know about AI research trends, from @natolambert Wednesday mornings weekly, sometimes extra posts.
Luke 🐝 Hewitt @lukebeehewitt
405 Followers 172 Following AI persuasion, manipulation and influence @TransluceAI Also @ai4pb_stanford
Kevin Lee @KLee4422
123 Followers 229 Following Member of Recruiting Staff at @LiquidAI my opinions are my own
Andy Coenen @_coenen
6K Followers 2K Following AI controllability @GoogleDeepMind | ex The M Machine | opinions my own
Anthony Aguirre @AnthonyNAguirre
4K Followers 154 Following CEO of Future of Life Institute. Physicist & cosmologist. Co-Founder of Future of Life Institute, Foundational Questions Institute, and Metaculus. Pro-human.
Agam Bhatia @chewing_a_gum
496 Followers 2K Following professional scriptrunner, CS + Math @Stanford
Institute for AI Poli... @iapsAI
2K Followers 62 Following Shaping AI’s impact on security, innovation & global affairs.
Axel Højmark @AxelHojmark
47 Followers 151 Following Research Scientist working on AI Safety at @apolloaisafety Previously: @MATSprogram
Arcee.ai @arcee_ai
15K Followers 446 Following Open models from an American research lab. https://t.co/6gcSwU5nk3
AI Notkilleveryoneism... @AISafetyMemes
168K Followers 1K Following Techno-optimist, but AGI is not like the other technologies. Step 1: make memes. Step 2: ??? Step 3: lower p(doom)
Kelsey Hightower @kelseyhightower
3K Followers 0 Following
Emma Steuer 🧚🤖 @emmysteuer
28K Followers 2K Following Software engineer. Inside the companies building the future. Deep tech, defense, energy, hardware. Host @MTSlive Prev @Planet
Florian Brand @xeophon
15K Followers 788 Following evals @PrimeIntellect | open models @interconnectsai
rachel @chaotictransfem
2K Followers 1K Following 🏳️⚧️ she/they, cool transfem and very bi | 21 | stuck on this hellsite | ai research scientist + cybersecurity + hobby photography | nerd @pangram
nielsrolf @nielsrolf1
198 Followers 741 Following my favorite type of circles are wide moral circles
Harry Mayne @HarryMayne5
477 Followers 1K Following OxRML & BOLD @uniofoxford. Prev. Astra at Truthful AI with @OwainEvans_UK
Jan Dubiński @jan_dubinski_
437 Followers 1K Following TruthfulAI | PhD in AI | Warsaw University of Technology and NASK, working on AI safety and generative models
Johannes Treutlein @j_treutlein
536 Followers 201 Following Research Scientist at Truthful AI, working on alignment in LLMs with @OwainEvans_UK. Formerly @AnthropicAI.
Thinking Machines @thinkymachines
179K Followers 1 Following Thinking, beeping, and booping. @tinkerapi
Mira Murati @miramurati
845K Followers 638 Following Now building @thinkymachines. Previously CTO @OpenAI
Mario Zechner @badlogicgames
65K Followers 1K Following Armin's handler at https://t.co/B05ybKGkzx. Old man yelling at Claudes. https://t.co/Q1wG57v1yc https://t.co/mnOoWUr0TO https://t.co/8i5vIRE0Wn
Trends for United States
You might like
