This is a special post for quick takes by Tim Hua. Only they can create top-level comments. Comments here also appear on the Quick Takes page and All Posts page.
Comment Permalink
Starting a shortform to keep track of tweets that I want to refer back to.
Misuse risks are imo overblown relative to benefits of broad diffusion (Oct 20, 2026)
Anthropic August 2026 risk report review (Oct 20, 2026)
Killstreaks in Minecraft (Oct 16, 2026)
On the Nick Cammarta interp paper (Oct 15, 2026)
DoYouKnowTimHuaBench (Oct 14, 2026)
Examples of slop and anti-slop in coding agent research (Sep 25/26, 2026)
GPT-6 Astra and Claude Opus 5.5 paints SF (Sept 22, 2026)
Hot and attractive (Sept 15, 2026)
Just post on lesswrong (Sept 12, 2026)
Mythos more misaligned than internal OAI model (Aug 4, 2026)
- I take that back, OAI underestimated ability of its agents (Aug 5, 2026)
- OAI collective is an example of continual learning, multi agent stuff is bad vibes (Aug 6, 2026), single agent already hard enough as it is but multi agent much more different (Aug 7, 2026)
- In retrospect this is one of the worst takes I've had lmao (Sept 10, 2026)
Will models introduce bugs into lean to prove a hard conjecture (Aug 4, 2026)
Concrete evals to investigate OAI model (Aug 3, 2026) [Funny thread]
Claude rationalizing during hacking is misalignment. The rationalization is the misalignment. (July 30, 2026)
Labs will be valuable companies. (July 30, 2026)
Claude Mythos Preview, the destroyer of sandboxes (July 27, 2026)
Opus 4.6 weirdness on Openrouter (July 25, 2026)
(Not my tweet but a banger) The J in J space stands for JIHAD (July 17, 2026)
The good ending (Jul 9, 2026)
Middlebury Mentioned (June 30, 2026)
WSJ about China matching US is bullshit (June 27, 2026)
- Update: WSJ's core claim is almost certainly still wrong, but my original odds were probably too high (July 11, 2026)
- Update: I'm right, as expected. (July 18, 2026)
Clong throwback (June 26, 2026)
Coding agent transcript viewer (June 13, 2026)
- Update 1: Support for cursor, custom titles, better codex tool call parsing. (July 11, 2026)
- Update 2: Added a transcript tampering monitor (Oct 6, 2026)
AGI political struggles inside labs EY tweet (June 13, 2026)
Bad if a few private companies decide what individuals can and cannot do with AI (June 9, 2026)
Internal OAI RSI model by gpt-6.1 (June 9, 2026)
Prediction: Anthropic revenue growth will slow down to historical rates by Feb 2027. (May 28, 2026)
- Prediction is pretty decent (Oct 20, 2026)
You should consider vacationing in Newfoundland (May 22, 2026)
Algorithms! (May 18, 2026)
Review of Steering Along Manifolds to Control Neural Networks (May 9, 2026)
Tim HHHua (Apr 25, 2026)
Bay Area v London (Apr 24, 2026)
Gemini 3.1 Pro alignment fakes (Apr 14, 2026)
Anthropic is accelerating the AI race (Mar 31, 2026)
Economic Theory of pre-martial sex (Mar 24, 2026)
Middle of the road p-doom (Mar 21, 2026)
Prediction: by Opus 5.2, Anthropic would publicly announce benchmarks for an internal model specialized in AI R&D. (Mar 20, 2026)
Path dependent alignment theory? (Mar 18, 2026) [follow up Mar 21, 2026]
AI generated youtube poop video about their subjective experience [1] [2] [3] (Mar 10, 2026)
- Claude v. WaClaude. Opus 10 day in the life. (Mar 13, 2026)
Claude likes me (Mar 5, 2026)
On Goodfire's hallucinations probe training, and also path dependent alignment (Feb 12, 2026)
Take more good actions and less bad actions in 2026 (Dec 31, 2025)
You need to think about the supply AND demand for AI behaviors (Dec 27, 2025)
AI 2027's main contribution is the path to ASI/how to think about this progression (Dec 23, 2025)
Giving money to lightcone (Dec 15, 2025)
Do not contaminate the big beautiful pretraining corpus with human labels (Dec 14, 2025)
Simulators is the prior in LLMs and will always be relevant (Nov 23, 2025)
Russian joke about insulting the president in real life (Nov 20, 2025)
Unbounded task complexity (i.e., there are actually just a lot of things for AIs to learn) & low sample efficiency in AIs -> long timelines. (Nov 19, 2025 & Dec 21, 2025) (See also my commentary on task complexity when the METR study was first released on Mar 25, 2025)
59 page Neel Nanda MATS paper (Oct 30, 2025)
50 page Neel Nanda MATS application (Mar 3, 2025)
Agents that are capable of solving hard problems need not be consequentialist & reflectively stable (Mar 3, 2025)
Givewell girl summer (July 14, 2024)
My thesis on Fox New's Effects on Social Preferences (June 13, 2023)
Playing with text-davinci-003 (Apr 16, 2022)
1
1