kaiwilliams

Karma: 142

kaiwilliams 11 Sep 2026 15:00 UTC
1 point
0
in reply to: Cleo Nardo’s comment on: strawberry calm’s Shortform
To add a +1 to this, I’m pretty high-context on AI, but didn’t really clock the significance of him working in pre-training as being more notable than just an “AI employee.”

kaiwilliams 9 Sep 2026 1:45 UTC
1 point
0
in reply to: tdo’s comment on: MarkelKori’s Shortform
To build on @Garrett Baker comment, there’s public evidence that it’s an Astra fork: https://xcancel.com/chi_t_williams/status/2097440947889954974#m.
OpenAI’s internal model that solved Navier Stokes was probably in training before August 28. That day, the company restarted its “large frontier RL run” that it had previously paused while dealing with the hugging face incident.
OpenAI claimed today that the Navier Stokes solution came from a swarm of AIs of a new internal model. They write “Since August 28 we have been training a new internal model” that’s a lot better than Astra
I think this partly because this seems to have been the company’s largest planned frontier RL run. Or at least, OpenAI mentioned pausing such a run on August 18th in its pacing announcement
It’s possible that this new internal model is distinct from the other large RL run. But given that the run started (or re-started) on the same day, and the model seems really good already, I think it’s probably the case that the Navier Stokes solving model isn’t that new

kaiwilliams 9 Sep 2026 1:22 UTC
59 points
23
in reply to: Thomas Kwa’s comment on: Thomas Kwa’s Shortform
Lord let me survive, or if I die, to die out of distribution

kaiwilliams 20 May 2026 19:12 UTC
2 points
0
in reply to: leogao’s comment on: leogao’s Shortform
I wonder if there’s a question-asking game, preferably one-on-one that would encourage this? Something akin to NYT’s 44 questions to make anyone fall in love, but instead 44 questions to stare into the abyss. Getting the right interlocutor and the right questions would be hard to do though.
It’s not a game, but it is a structured activity.
I’m skeptical that you can really get the abyss in small doses. Maybe there’s also a progressive activity where the first exercises are small things to admit about oneself, before progressing to more and more difficult questions.

kaiwilliams 15 Apr 2026 21:45 UTC
10 points
0
on: kaiwilliams’s Shortform
Does anyone have thoughts on Muse Spark that they’ve written up? Do we have any speculations on what its good at/it’s size/whether it has good/bad post-training?

kaiwilliams 27 Mar 2026 15:39 UTC
3 points
0
in reply to: leogao’s comment on: leogao’s Shortform
The default advice blogging advice I’ve heard is that “obvious” topics often make for good posts because they are often non-obvious to readers, so one should strongly default towards posting if the concern that it is too obvious.
But maybe you’re making this judgement even with that prior in mind? I’d be curious to see one of these “obvious” posts.

kaiwilliams 15 Mar 2026 22:36 UTC
2 points
2
on: New LessWrong Editor! (Also, an update to our LLM policy.)
I realize that part of the goal is to make the LLM portions unobtrusive, but would it be possible to make LLM sections have a collapse button at the top? (Or bottom). By default they can be open.
When reading current LessWrong posts that have LLM sections, I find myself mostly skipping LLM sections and appreciate when someone has placed them in a collapsible.

kaiwilliams 27 Feb 2026 15:12 UTC
1 point
0
in reply to: Edd Schneider’s comment on: Life at the Frontlines of Demographic Collapse
I was going to comment this about Vermont too, though perhaps a bit less intense than upstate NY. One of the culture shocks of moving away from VT for college was realizing that not many other Americans had faced the feeling that the way of life was ending.

kaiwilliams 17 Feb 2026 21:59 UTC
2 points
0
on: models have some pretty funny attractor states
Cool work! I would be curious to see what attractor states exist between models? What happens if you plug up a Gemini model with a Claude model, e.g.?

kaiwilliams 26 Jan 2026 18:25 UTC
16 points
2
on: The Possessed Machines (summary)
I was just dragged through Demons for a book club, so I was amused to read this. At least it means the time I spent reading that wasn’t in vain.
There’s some stuff that feels a little bit weird here. The author says they left in early 2024 and then spent the “following months” reading Dostoevsky and writing this essay. Was the essay a bit older and only got put up? (Has to be relatively recently edited, if it was run through 4.5). Who are the editors alluded to at the very end? Is it supposed to be Tim Hwang? A little bit more transparency would be much appreciated (the disclaimer about Opus 4.5 being used for anonymization was only added on the 24th after some people had pointed out that it sounded rather AI-written.).
Another weirdness: why did Hwang put up another microsite about Demons that’s written by an anonymous author “still working in industry” that has clear LLM-writing patterns at basically the same time? https://shigalyovism.com/. Though this one is much less in-depth.
Can anyone with more experience in the frontier labs/the uniparty give a sanity check for whether this seems like it was written by someone who is who they say they are?
What links here?
- Problems with “The Possessed Machines” by Eye You (29 Jan 2026 21:00 UTC; 34 points)
- Eye You's comment on The Possessed Machines (summary) by L Rudolf L (29 Jan 2026 21:08 UTC; 7 points)

kaiwilliams 23 Jan 2026 18:22 UTC
3 points
0
in reply to: dynomight’s comment on: dynomight’s Shortform
Ah right right—I remember reading that post. The subscribe form using dynomiiiiiiiiiight makes sense, especially given how I prompted Llama: I pasted the post in and then appended Author:
I am curious if there’s a way to get an instruction tuned model to role play being a base model, and see if they do better at truesight than regular instruction tuned models. Like, why do chat models get worse? Is it that the assistant character is bad at that? Plenty of interesting questions here.

kaiwilliams 23 Jan 2026 16:58 UTC
6 points
0
in reply to: dynomight’s comment on: dynomight’s Shortform
Llama 3.1 405B base: dynomiiiiiiiiiight
I resampled it a couple times and it added a couple of i’s to your handle consistently (despite getting your url dynomight.net, so it clearly knows you). Not quite sure why. Weird that base models are so much better at this.

kaiwilliams 8 Jan 2026 0:22 UTC
3 points
0
on: Festival Stats 2025
Thanks for compiling! It feels apt that the name of the top caller is Will Mentor.

kaiwilliams 7 Jan 2026 2:56 UTC
1 point
0
in reply to: AndrewS’s comment on: AndrewS’s Shortform
Do Lesswrong quick takes count as social media? :)

kaiwilliams 1 Jan 2026 22:59 UTC
1 point
−1
on: Recent LLMs can do 2-hop and 3-hop latent (no-CoT) reasoning on natural facts
On this dataset, I find that Gemini 3 Pro gets 60% of 2-hop questions right and 34% of 3-hop questions right.
I initially got tripped up by the wording here: I thought this was 60% accuracy on 2-hop questions in a forward pass, not with 300 filler tokens, which aren’t mentioned until later in the post.
It’s a good piece, but wanted to comment in case someone else gets confused at the same spot.

kaiwilliams 1 Jan 2026 22:55 UTC
5 points
0
in reply to: Adele Lopez’s comment on: Recent LLMs can do 2-hop and 3-hop latent (no-CoT) reasoning on natural facts
I came here to comment the exact same thing. I wonder if 2-hop latent reasoning is correlated well with Simple-QA scores.

kaiwilliams 11 Dec 2025 18:43 UTC
1 point
0
in reply to: technicalities’s comment on: AI in 2025: gestalt
Kudos to Deepmind for being the first to release output watermarking and a semi-public detector. You can nominally sign up for it here.
Afaict, some of this is now in the Gemini app. But if not, feel free to ping me (I have access).

kaiwilliams 3 Dec 2025 1:40 UTC
3 points
0
on: Unless its governance changes, Anthropic is untrustworthy
The only public instance of this change being pointed out was a LessWrong comment by someone unaffiliated with Anthropic.
Nitpick: an outside reporter also noticed this on the day of the release and wrote up a story on it. It didn’t seem to get much traction though.

kaiwilliams 17 Oct 2025 16:46 UTC
5 points
2
in reply to: eggsyntax’s comment on: eggsyntax’s Shortform
I thought the “past chats” feature was a tool to look at previous chats, which only happens if the user asks for it, basically. (I.e., there wasn’t a change to the system prompt). So I’m a bit surprised that it seemed to make a difference around sycophancy for you? But maybe I’m misunderstanding something.

kaiwilliams 3 Oct 2025 13:53 UTC
1 point
0
in reply to: FVelde’s comment on: No, That’s Not What the Flight Costs
Labor costs are much higher in the US, which I think plays into this. So it’s easier in Europe to not be reliant on the credit card model.