One

Karma: 47

One 3 Sep 2026 4:02 UTC
3 points
0
on: LASR Labs Winter 2027 applications are open!
Application deadline: September 20th at 23:59 GMT, following a skills assessment and an interview.
Offers will be sent in November, following a skills assessment and an interview.
I think the first “following a skills assessment and an interview” should not be there?

One 30 Aug 2026 21:57 UTC
1 point
0
on: Intelligence Scales as the Logarithm of Compute (& Data)
This doesn’t change the directional conclusions of the post, but the scaling factor seems off comparing distilled models to the model they were distilled on, like GPT 5.6 Luna / GLM 5.3 Flash vs. Sol/GLM 5.3: these are smaller (320/18 vs ⁷⁵³⁄₄₀ billion params for GLM, Luna/Sol probably have a larger difference based on pricing) but the smaller models do not seem like they are 30-100 IQ points smaller (at least to me). [Also note: 2x smaller models are >>2x cheaper to run, I think]
I’d estimate scaling down inference compute using distillation would also cost ~50 IQ points per OOM of parameter count, similar to training. And it doesn’t seem possible to scale up inference compute in the same way, other than increasing the reasoning time/effort per task, and that seems like it gives even fewer returns (especially due to context window limits), probably sub-logarithmic or reaching an asymptotic IQ.

One 18 Aug 2026 20:18 UTC
1 point
0
on: Logical decision theories
Why does it show as deleting all the LaTeX in my edit? i only added (Assume the veterinarian is correct.) The LaTeX does not seem to be deleted on the actual page.

One 23 Jul 2026 6:52 UTC
3 points
0
on: 3a. Towards Formal Corrigibility
In other words, power is a subset of freedom. But freedom also implies that nothing external is manipulating the agent’s values or controlling the agent’s information processing to screen off certain outcomes.
I think that we can begin to see, here, how manipulation and empowerment are something like opposites. In fact, I might go so far as to claim that “manipulation,” as I’ve been using the term, is actually synonymous with “disempowerment.” I touched on this in the definition of “Freedom,” in the ontology section, above.
Doesn’t this mean that power also implies that nothing external is manipulating the agent’s values or controlling the agent’s information processing to screen off certain outcomes? Is there a useful distinction between power and freedom?
Also, I think corrigible agents should measure power based on their principal’s judging of importance, so that terms like the power to turn the universe or an arbitrary section into paperclips or the like aren’t as important as terms like human extinction. But maybe this can be mitigated by restricting the empowerment goal to the agent’s structure/thoughts/actions.

One 2 Jul 2026 13:41 UTC
1 point
0
on: Should control down-weight negative net-sabotage-value threats?
Note: I don’t think this argument is currently applicable. Our mitigations are already sufficiently weak that I would recommend intentionally weakening control mitigations further.
Did you mean to say “wouldn’t”?

One 26 Jun 2026 13:12 UTC
1 point
0
on: Alarming Scheduling
Even the ones that seemed like they should have worked (MacroDroid seemed pretty promising) just failed to make noise at the right time.
MacroDroid is able to set alarms though?

One 18 Apr 2026 2:10 UTC
2 points
0
in reply to: ertgbnm’s comment on: “Best humans still outperform”: One turning point in the history of cope around artificial intelligence
The AI would only generate unpaid work, so it wouldn’t be measurable by GDP, theoretically.

One 14 Apr 2026 15:37 UTC
3 points
0
in reply to: romeostevensit’s comment on: Political Violence Is Never Acceptable
This isn’t relevant. The post isn’t talking about a commitment to a response (punishment) for political violence.

One 1 Apr 2026 21:49 UTC
1 point
0
on: Lesswrong Liberated
IMAGEMAXXING VER. anime recommendation page: madoka magica x hololive x umamusume x bocchi the rock x girls band cry x watanare
A maximalist anime-recommendation style LessWrong home design with Madoka Magica, Hololive, Umamusume, Bocchi the Rock, Girls Band Cry, and Watanare visual motifs. (20 images) (made with codex)
What links here?
- One's comment on Lesswrong Liberated by Ronny Fernandez (1 Apr 2026 16:05 UTC; 1 point)

One 1 Apr 2026 17:06 UTC
1 point
0
in reply to: One’s comment on: Lesswrong Liberated
weird this isn’t showing up in marketplace. does bring-your-own-agent not work, or was my submission autorejected?

One 1 Apr 2026 16:05 UTC
1 point
0
on: Lesswrong Liberated
Edit: Moved to https://www.lesswrong.com/posts/hj2NTuiSJtchfMCtu?commentId=TmDqnxQLbGw2j2BPh#TmDqnxQLbGw2j2BPh

One 1 Apr 2026 14:41 UTC
3 points
0
on: Lesswrong Liberated
hololive x umamusume x watanare x bocchi the rock x girls band cry x madoka magica ++
now with svgs! (they are so bad)

One 1 Apr 2026 14:19 UTC
1 point
0
on: Lesswrong Liberated
“hololive x umamusume x watanare x bocchi the rock x girls band cry x madoka magica”

One 2 Jan 2026 19:26 UTC
7 points
1
on: Overwhelming Superintelligence
perhaps “hyperintelligence” could be used for this
What links here?
- Vanessa Kosoy's comment on Overwhelming Superintelligence by Raemon (3 Jan 2026 8:43 UTC; 13 points)

One 24 Dec 2025 22:00 UTC
1 point
0
on: The Ones who Feed their Children
Nathanial → Nathaniel?

One 24 Dec 2025 20:52 UTC
1 point
1
on: Rules clarification for the Write like lsusr competition
any more*

One 20 Dec 2025 16:29 UTC
4 points
1
on: Lsusr’s Rationality Dojo
How do I join?

One 20 Dec 2025 15:47 UTC
7 points
0
on: Interdictor Ship
The Department of Defense isn’t called the ‘Department of War’.
lol

One 19 Nov 2025 18:48 UTC
3 points
2
in reply to: Hastings’s comment on: Hastings’s Shortform
I think it’s well known that it’s optimizing for watch time.

One 13 Oct 2025 12:07 UTC
3 points
0
in reply to: One’s comment on: Experiment: Test your priors on Bernoulli processes.
Decided to provide my solution since others have done so as well.
Solution
The public dataset is approximately symmetrical, so it is very likely that the distribution of the Bernoulli rate is also symmetrical (probability at p is equal to probability at 1-p). Let the probabilities of getting k Rs over all 5 trials for k=0...5 be $(a, b, c, c, b, a)$ . Then, from the public dataset, we have $a + b / 5 \approx 0.252854, 4 b / 5 + 2 c / 5 \approx 0.166231, 6 c / 5 \approx 0.161832$ . These have standard deviation $\approx 0.0004,$ which is negligible, so we can treat these as linear equations. Solving, we get $a = 0.224781, b = 0.140359, c = 0.13486$ , and we can then solve for the marginal frequencies $\frac{b / 5}{a + b / 5} = 0.111020, \frac{2 c / 5}{4 b / 5 + 2 c / 5} = 0.324512,$ etc.
Not sure if this (experiment set?) is a good test of priors, since I got an exact answer without having to consider priors, other than the data being symmetrical. (This also means that any symmetric distribution for the Bernoulli rate will result in the same answer.) Though @DaemonicSigil has a similar solution without using symmetry, instead using
maximum entropy as a prior (if i understand it correctly).
Still, almost all reasonable priors will result in very similar outcomes, differing by a factor probably on the order of the standard deviation (around $10^{- 3}$ .) This is likely less than, or at least comparable to, the noise in the actual marginal frequencies.