Adam Scherlis

Physicist → AI safety researcher @d_model_ai

[Adam Scherlis @ascherlis]

@MoleculesGoHard I am now distractingly tempted to find a pair of valid codes related by GoL evolution

[Adam Scherlis @ascherlis]

@allTheYud And "controversial" seems to mean "probably bullshit"

[Adam Scherlis @ascherlis]

@allTheYud Known processes cost much more than $1/J; that's the cost of energy (due to inefficiency) alone.

[Adam Scherlis @ascherlis]

@allTheYud Based on Wikipedia alone, I would bet against "emittable as [stimulated] gamma radiation" being true, at even odds.

[Adam Scherlis @ascherlis]

@TetraspaceWest just like me fr

[Adam Scherlis @ascherlis]

@benwr pretty much?

[Adam Scherlis @ascherlis]

@AndyMasley It's also frustrating to discuss because I don't actually think "[future] AIs are just like people" is better than "AIs are just tools". They're a secret third thing! There are about to be three kinds of things in the world (or four if you count nonhuman animals).

[Adam Scherlis @ascherlis]

@AndyMasley This manifests in a bunch of different ways, but you can see the underlying connections when people jump from arguing against the prospect of AI agency/goal-seeking to AI consciousness to AI welfare and so on.

AIs having interests of their own to pursue seems to be a big leap.

[Adam Scherlis @ascherlis]

@_bbctol I have several long-beaten tests (in ~2020, ~2022, and ~2024 respectively), and a very qualitative unbeaten one.

[Adam Scherlis @ascherlis]

The spoilsports may have also been Einstein, I'm not sure.

[Adam Scherlis @ascherlis]

This is literally how Special Relativity works and why light "doesn't experience the passage of time".

Einstein even used "i c t" as his fourth spacetime coordinate, with a Euclidean metric.

Later, some spoilsports decided to absorb the i² = -1 into the metric.

[Math Lady Hazel 🇦🇷 @mathladyhazel]

I can’t stop laughing.

[Adam Scherlis @ascherlis]

@MiltonEverglade @Lewiswbt1 @shaddytheguy https://t.co/1ZB6Htq3if
dammit

[Joseph Laposata @demhockracy]

@Lewiswbt1 @nikicaga One must imagine Simpsonphus happy

[Adam Scherlis @ascherlis]

@MiltonEverglade @Lewiswbt1 @shaddytheguy We must imagine Simpsyphus happy

[Adam Scherlis @ascherlis]

@ScienceNews2day @mathandcobb No, the faint gray grid lines are a unit apart (note the axis labels). Nearly all of the graph edges run through a couple extra vertices in the middle. Around the margins you can find some edges that don't get interrupted.

[Adam Scherlis @ascherlis]

@mathandcobb (I don't understand the problem well enough to know if disc vs square actually matters, mind you)

[Adam Scherlis @ascherlis]

@mathandcobb You should double-check the "polydisc" part of the construction btw, the projection from the ring looks right to me but I'm not sure about that part.

[Peter Schmidt-Nielsen @ptrschmdtnlsn]

There are several results that morally point in the direction of "you should expect endless rich beauty", but my favorite is this:

You might naively worry that at some point we prove everything that's easy to describe, and we have to make up extremely complicated questions. And at that point, is it really that interesting to say that some 50 page contrived problem required some complicated theory to prove?

However, we know this won't happen! If you take all questions of n characters, take the shortest proof of each, then look at the growth rate of the length of the longest such shortest proof, you find that the length must grow uncomputably quickly in n. Otherwise theorem proving would be computable by proof search, which would let you decide provability, thus a fortiori decide the halting problem.

So, we must have a vast supply of very simple questions whose answers are spectacularly complicated. In practice, this is what we see, and the complexity seems to correspond to real richness!

For example, as near as we know, the answer to "when does x^n+y^n=z^n have solutions over Z?" is just massively incompressibly deep, requiring the development of extremely sophisticated tools. Likewise, "what can be said of a group that's finite and simple?" seems to just be a massively deep question. That simple questions can require thousands of pages of deep theory should be unsurprising in light of this "proof lengths must grow uncomputably quickly" result!

And "grows uncomputably quickly" is an absolutely staggering growth rate. There is likely some short couple-paragraph question where resolving it would require you to develop one million pages of rich theory, beyond the intellect of any human.

[Adam Scherlis @ascherlis]

@conundrumer @pleometric Cairo tiling 👀

[Adam Scherlis @ascherlis]

Not very precise phrasing on my part.

It's *constructive* but not *an explicit construction*, in the sense of having lots of steps like "there's clearly many options here and we can show that at least one of them will work".

But the paper doesn't actually figure out which options are the ones that work at each step, for various n, in order to get a concrete sequence of graphs. It gives a recipe for a construction but doesn't carry it out. Which is sensible because the resulting graphs (even in Sawin's heavily-optimized version) would be astronomically large.

I didn't mean to imply the proof was nonconstructive in a proof-theoretic sense.