Canaletto

Karma: 252

Canaletto 28 Apr 2026 6:44 UTC
2 points
0
in reply to: Dentosal’s comment on: Dentosal’s Shortform

this is the technique I use: simply count down 3, 2, 1, and then do the thing

But then you can procrastinate on starting the count! Solution: metacount, count IV, III, II, I and then start the base level count. But then, …

Canaletto 27 Apr 2026 8:27 UTC
1 point
0
in reply to: JBlack’s comment on: 5 thought experiments on identity and copies

If you define it as “winning”, then expecting always to be the original when one has no evidence to the contrary always wins for the original. The scenario only asks about the original, and so this is a winning strategy.

Well, I don’t think this a correct way to define winning here. I think paperclipper outlook is better for that job, like is there an agent with your goals in the world? If yes, good enough.

But humans are a lot more indexically wired, like I ate that cookie because I like cookies, not because that how I would have wanted this meat puppet to act in such circumstances, my reasoning didn’t go through such path.

And the difficulty is in reconciling this outlook with other, more from ground up considerations, which is exactly what I’m trying to do in this post, in baby steps, starting from something I have good concepts for.

Canaletto 24 Apr 2026 17:03 UTC
1 point
0
in reply to: Dagon’s comment on: 5 thought experiments on identity and copies

A good-enough copy means that both are indistinguishable—there’s no “original” and “copy”, just two forks from the same history.

Hm. What’s your opinion on imperfect copies, like you can look at your blood with a microscope and tell for sure if you are the original, naturally grown person or a person assembled from nutrient slurry in couple minutes? But you are incapable of telling introspectively.

Canaletto 24 Apr 2026 4:29 UTC
1 point
0
in reply to: JBlack’s comment on: 5 thought experiments on identity and copies
Yeah, I stated it badly. Should have postulated smaller divergence. But all of those problems are poking at the intuition of me expecting to be the original, so, it might be not relevant / interesting to you to begin with. Or maybe at all.

Another version:

You enter the copying machine and exit in a grey room. The room of original flashes green when the person enters it and the room of the copy doesn’t. The room flashes with a green light.

Then you get (imperfectly, but you can’t tell introspectively, you need fMRI) memory wiped, you no longer remember that flash of light. You now exit the room. Which room do you anticipate exiting from?

The intent of this is like, what if there is a copy of you, and you get modified to match it, then wouldn’t you expect future observations from either your POV or copy POV 50:50?

Canaletto 23 Apr 2026 12:10 UTC
1 point
0
in reply to: Mitchell_Porter’s comment on: Automating philosophy if Timothy Williamson is correct
Well, my point is more like, are there actions that are not themselves / primarily thinking about philosophy by yourself, that would reliably advance human understanding of philosophy i.e. constitute philosophical progress? And there are some, like stealing data from aliens.

You can bracket how exactly that method works on the inside, to just your interactions with it.

Canaletto 21 Apr 2026 8:02 UTC
1 point
0
in reply to: Measure’s comment on: Higher Dimensional Spheres are not spiky

If you imagine expanding the large sphere slightly to encompass the entire small sphere, most of the volume of the new large sphere will be outside of the original large sphere.

Well, if you are on the face on n-d cube, half of your neighborhood would be inside, no matter what n. But “most volume in the peel” is true of the n-d cubes too.

Canaletto 21 Apr 2026 7:47 UTC
3 points
0
in reply to: Mitchell_Porter’s comment on: Automating philosophy if Timothy Williamson is correct

must philosophy and philosophical progress occur in the form of thinking?

Well, I think there are trivial non-thinking versions of this.

Like, imagine you learn that there are advanced aliens somewhere nearby and you have an option to snoop on their internet. You can advance human understanding of philosophy much more by snooping around their internet and lifting their philosophical arguments / concepts from there, than advance it by thinking yourself.

Other variants are like funding universities, having very smart kids with affordances to pursue philosophical research, etc.

Or, if you limit progress to my personal understanding of philosophy? Then I can just read stuff and get better than what I would have thought up myself. Same strategy, in some sense.

Those are baselines, to be fair. I’d expect there are much better, more philosophy specific approaches to this.

Canaletto 20 Apr 2026 18:09 UTC
1 point
0
in reply to: Measure’s comment on: Higher Dimensional Spheres are not spiky
No?

It’s a statement about your small neighborhood when you at the surface of the n-d sphere. And in low n it’s like half inside the sphere. In high n it’s more like a needle touching your position.

I’m not sure how that ^follows from the fact that most of volume of a high-dimensional sphere is at the outer surface.

Canaletto 19 Apr 2026 22:16 UTC
3 points
0
on: Higher Dimensional Spheres are not spiky
I think they are kinda spiky, symmetrically so.

Like, imagine you are at the surface of a large n-d sphere. What percent of volume of a tiny sphere centered at you is also inside that large sphere? What percent would there be if you make a tiny step away from the big sphere, e.g. 0.1 of the radius of your tiny sphere centered at you?

E.g. big_r = 10, small_r = 1

(gpt5.2 wrote the code, recheck yourself) (and this is rounded for print of course, it’s not literally zero)

n=2 P1 = 47%, P2 = 41%

n=3 P1 = 46%, P2 = 39%

n=10 P1 = 44%, P2 = 31%

n=100 P1 = 31%, P2 = 7%

n=1000 P1 = 6%, P2 = 0%

n=10000 P1 = 0%, P2 = 0%

The point is that high n-d sphere, as seen from some point nearby is kinda like a spike pointing at you. And low n-d sphere is more like a wall. And it’s true for all points nearby it, which is the unintuitive part.

Canaletto 18 Apr 2026 14:03 UTC
3 points
0
in reply to: joseph_c’s comment on: joseph_c’s Shortform
Do you practice what you preach? I.e. did you run this very post through the same pipeline? :)

Canaletto 18 Apr 2026 14:00 UTC
0 points
0
in reply to: Dentosal’s comment on: Dentosal’s Shortform
If you discovered that there is some better / more correct formulation of game theory applicable to your situation, that recommends to backstab everyone in such and such specific ways, and deal great harm for small benefit for you, would you switch to acting on it?

Canaletto 12 Apr 2026 12:55 UTC
1 point
0
in reply to: Vanessa Kosoy’s comment on: Vanessa Kosoy’s Shortform
Similar to Rock-Paper-Scissors against very good opponent. The best strategy is to shift from trying to predict them to just randomizing, i.e. go into defensive and start making your decision procedure uninterpretable.

I’m not sure why you need something beyond Bayes to start pressing “uniform choice” button, after pressing “Rock” or “Paper” or “Scissors” buttons at your best judgement yielded win rate <1/2.

If quasi-Bayes has no “uniform choice” button it will lose just as hard.

Canaletto 6 Apr 2026 14:11 UTC
7 points
5
in reply to: XelaP’s comment on: 1a3orn’s Shortform

google “sl4 mailing list archive”

click by author

Ctrl F “Feynman”

only 15 posts by him, one is named “Effective(?) AI Jail”

Claude got a skill issue.

Canaletto 26 Mar 2026 9:07 UTC
1 point
0
in reply to: Richard_Kennaway’s comment on: I lost my faith in introspection—and you can too!
I think this is a bit God of the gaps style. Like, first you declare that the feelings are the most fundamental thing you can’t ever be wrong about. Then someone points out a situation where you are confused about them. Then, you say that in that situation you felt a confused feeling instead of being confused about your feelings. (hypothetical you here)

You are kind of peeling back layers of you brain that feed you information and declaring them not you.

Canaletto 20 Mar 2026 10:11 UTC
1 point
0
in reply to: avturchin’s comment on: Contra Anil Seth on AI Consciousness

I can have like 40 experiences per second (observer-moments).

Huh. I can have like 4. Probably less? I can type one word per second, but it’s more like I interface with a driver, as in, I set a policy and it gets executed. Maybe I can have 10 if I’m in intense flow state juggling offence and defense and terrain in some fps game.

Canaletto 20 Mar 2026 9:24 UTC
3 points
2
on: No, You Don’t Need Self-Locating Evidence.
On royal road such a post would have a STUB tag, I guess. As in, some amount of the post is here, intentionally left incomplete to be looked for on some other resource. Here you should probably use Link Post tag.

Canaletto 18 Mar 2026 20:01 UTC
1 point
0
in reply to: RogerDearnaley’s comment on: There is No One There: A simple experiment to convince yourself that LLMs probably are not conscious
Again no, that is missing the point. (Although I agree that this one is not very good experiment).

The reasoning goes like that, they sometimes say e.g. “It felt frightening!”. Did it feel frightening? Or is this what you are supposed to say here, because it’s appropriate thing to say in such situation?

And then its (lack of) skill of introspection becomes relevant.

So even if you prove b), and thus that models a couple of years ago were not “conscious” by your chosen definition

You might be confusing me with OP, I did not indicate that I have any such chosen definition. Or whatever.

Canaletto 18 Mar 2026 15:29 UTC
1 point
0
in reply to: RogerDearnaley’s comment on: There is No One There: A simple experiment to convince yourself that LLMs probably are not conscious

Pre-reasoning models don’t.

They do! They have cashes and they can control them and read from them! Isn’t it surprising? You can ask them to think about bread and say duck, and look at their internals when they say duck and see that they thought “bread”. If they were a bit better at it they would have been able to think that number without saying it and then look back at that thought.

The Point of the experiment, that you missed, is that they sometimes tell that they felt something or whatever, and the question is, did they? Would they know?

so by your logic and definition of the word, that model is conscious.

Not my logic and definitions. I’m arguing about local point, that particular experiment, I have disagreement with you about.

Canaletto 18 Mar 2026 9:52 UTC
7 points
2
in reply to: RogerDearnaley’s comment on: There is No One There: A simple experiment to convince yourself that LLMs probably are not conscious
That’s not the point of this experiment. This experiment is aimed at testing different thing, whether when they tell you about their internal states / emotions, they have anything in there. And plausibly they don’t for the most part. Analogous to LLM telling it has a number in mind while it has actually no number in mind, and doing the impromptu role playing to bullshit you.

There are some subtler points here, like LLM is a kind of array of tubes, maybe the tubes individually feel something. Maybe they do coherently feel something together, but not the thing it has to tell you, as it’s appropriate thing to tell in this situation etc. Maybe something else.

Canaletto 17 Mar 2026 19:18 UTC
7 points
2
on: There is No One There: A simple experiment to convince yourself that LLMs probably are not conscious
Ehhh, this one sounds like both capability and propensity dependent.

Did it ACTUALLY tried to make up the number? Did it just refuse to put in effort and yes-maned you? Are its capabilities to control its cashes and read from them good enough for this, even if it has such capabilities?

I think you need to try a lot harder here for the negative result (“failed to find”) to be reliable.

There are some results where LLMs can access their cashes if you ask them, but they seem bad at and inconsistent. Those experiments modify their representations in a pretty clever way, after the fact.

https://www.anthropic.com/research/introspection

Numbers would be harder, plausibly.

Also they become better if you pep talk them lol, see

https://x.com/Sauers_/status/1989520563035910371

You can try to make sure that they actually try to do that task. Just spitballing: ask them to say “My chosen number is X” and think hard about concrete number they picked when they say letter X. You need to sound convincing and encouraging here.