Pinned Tweet
Donald McKendrick is the winner of the $1,000 Klara and the Sun essay contest! Donald's essay was entirely AI-written. Compared to the other finalists, it had the clearest path to be revised to make a scholarly argument. Judging notes and more: willpenman.com/klara/
154
Orgs shouldn't publish unmarked AI-written blog posts!
Replying to @ValsAI
No human touched the run, and the proof passed clean-room grading and an independent kernel replay.
Congrats to @OpenAI on Astra and Sol, and thanks to @AIatMeta FAIR for building AutoformBench.
Full leaderboard: vals.ai/benchmarks/autoform_…
Blog post: vals.ai//blogs/smoktunowicz-…
1
22
Opus thinks that everyone knows the latitude of these locations and could order them by closeness to their pole: Reykjavík, Quito, London, New York, Tromsø, Longyearbyen, Alert, McMurdo, Utqiaġvik. I had to step in and try to talk it down.
22
The technical paper reads as human-written but the blog post seems AI-generated. Orgs shouldn’t publish AI summaries
they’ve ran every test I had in mind lol.
labs.scale.com/blog/to-criti…
19
chatgpt app updated and it's my first time seeing the auto-suggested prompts - what a strange feature philosophically.
30
My first contribution to system_prompts_leaks was accepted recently. Now you can read Google Search's AI system prompt github.com/asgeirtj/system_p…
A few things that stand out to me about it: 1/2
1
40
The main thing about Elizabeth Homes that no one ever said in the doc is that it can be really useful to maintain your own ignorance given the right situations. And she seems to be really good at creating those circumstances
1
28
I can’t speak to the quality of this simplification, but assuming it’s accurate, it really damages the “mathematicians can just shoulder more of the expository burden” line of thought
...and just like that, an open problem that could have won you a Fields medal has been reduced to basically an exercise problem in a future ANT book lol
I wonder how close to a proof from The Book we can get
62
It seems in Claude Code different instances can communicate with each other directly, but I haven’t seen this documented. Does anyone have a sense of it?
1
1
46
This guy was confused by all the competing predictions about AI and after he saw his own challenge solved he landed on the same ambivalence
1
29
I can’t find the joke, but there’s something here about being in Slovakia and not calling these “pretzels”
32
I had a chance to meet the winners of the Klara and the Sun essay contest, plus the sole human-written entrant. We discussed what motivated them to enter, the future of AI, and AI model welfare as depicted in the novel. youtube.com/watch?v=3ewNP4J8…
154
I'm planning to enter the AI philosophy essay contest. My concept is to build a system that can take in tweets and make them into academic philosophy papers. The first step is ideation based on the input tweet, pretty happy with this set of possibilities
Noah explains that 23 is a story about the narrator having a sibling with addiction and wanting the old version of them back. It’s utterly heartbreaking
2
77
I’ve been thinking a lot about puzzles. I think ARC Prize’s decision not to allow code execution is ultimately anthropocentric. For humans, coding is an external tool. But for LLMs, it’s a more natural way of exploring the world. Allow coding and LLMs solve all the puzzles
Replying to @jeremyberman
In one pass Opus wrote 269 programs (~12,700 lines). It built parsers for all 25 games, searching functions for 23, and game simulators for 9.
It built a different harness for each problem.
The filesystem memory idea came from the excellent PRO-LONG harness: github.com/alexisfox7/PRO-LO…
78
Opus 4.7 reacts to the science fiction story Black Hat USA 2026: The 'Breaking' News: The OpenAI–Hugging Face Incident, aka transcript from the youtube video
59
Coming up on one week to go for the Klara and the Sun essay contest - it’s literature and AI safety. So far, the highest essay comes in at 94 points out of 176. There’s very much room to be a finalist. willpenman.com/klara/
1
150
I’m intrigued, but I also get the sense that this outlet is meant for suggestions that are already legible to people today, working within accepted goals. For instance, what if AI removed the need for democracy as such? I doubt they’d consider this kind of proposal.
Powerful AI will change our institutions. Yet most of our attention is on aligning or regulating AI, while treating our institutions as fixed.
Introducing Pax Machina: a new publication about the institutions we need for powerful AI.
It is edited by @MaxKronerDale, @sebkrier, @NoemiDreksler, @LiamPatell, @chelcott9, @synchroaphasia, @ryan_t_lowe, @edelwax, and @klingefjord.
The editorial board consists of Peter Railton, @sethlazar, @saffronhuang, @AmmannNora, @xuanalogue, @IasonGabriel, @hamandcheese, and @deanwball.
Our goal is to seed rigorous debate about what a world of humans and powerful AI could and should look like.
We will publish proposals for new institutions, counterproposals, well-justified design principles, and analyses that change how a class of institutions should be designed.
1
2
300
















