iiinova — where your digital self keeps evolving

Computational Cosmetologist @dferrer.bsky.social · 8h

Which is fine. Score matching is trivial to defeat and fingerprinting can only be done by the vendor. They decided to rely on a trained classifier. But then they can't act like they still have those precision guarantees.

010

Computational Cosmetologist @dferrer.bsky.social · 8h

It's also infuriating because there are high-precision (but low recall) methods for AI authorship testing. If you have a fingerprint or find a model that score-matches you really do have near-certainty that it came from an LLM if the passage is long enough. Pangram doesn't do that.

110

Reposted by Computational Cosmetologist

ghost malone ergo propter bones @proptermalone.bsky.social · 16h

i think we probably do need to do more talking about how to help women heal from sexual assault and less talking about whether and how sexual-assualting men can or should be rehabilitated, based on that realization that i've seen a great deal of discourse on the latter and ~none on the former

3255174

Computational Cosmetologist @dferrer.bsky.social · 8h

There are a bunch of societal reasons *scholars* will tell you, but at least half of it is that cats are fucking hard to draw! They have weird flexible skeletons that make them look different in every pose / angle. Unless you can get one to sit still, you're in for a rough time without photography.

020

Computational Cosmetologist @dferrer.bsky.social · 9h

My prior can do literally anything on its now-finite support. There's no need for it to decrease at all. Why should it?

021

Computational Cosmetologist @dferrer.bsky.social · 9h

So, if I have a prior on the number of bits of the minimal program necessary to produce some output, for it to be a proper prior it must be decreasing in the limit. Great. But the universe is (probably) discrete and finite. So there is a longest program, and I should assign p = 0 beyond. Uh oh.

120

Computational Cosmetologist @dferrer.bsky.social · 02/10/2026

On the other hand, if you have anxiety about being underdressed, wearing a suit to a wedding is basically always fine unless the invitation specified otherwise. If the invitation doesn’t say anything though id expect dressed up or down to be fine. Just match your date / SO if you bring one

020

Reposted by Computational Cosmetologist

Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 30/09/2026

In our Nature paper, we introduce the first superhuman Stratego AI, which we built using general techniques that we developed for RL & test-time compute under imperfect information. www.nature.com/articles/s41...

nature.com

Scalable decision-making for games of imperfect information - Nature

Ataraxos, an AI for the board wargame Stratego, establishes a design pattern for reinforcement learning and search that is effective under large amounts of hidden information, a longstanding desiderat...

1225961

Computational Cosmetologist @dferrer.bsky.social · 01/10/2026

“Inconsistent reasoning about probabilities in limits” is the generating function to the Julia set that is the fractal wrongness of the Rationalist worldview

0201

Reposted by Computational Cosmetologist

rev. howard arson @theophite.bsky.social · 01/10/2026

somehow, despite the NFL theorem not applying to the real world, they have constructed a model of statistics which the NFL forbids

31448

Computational Cosmetologist @dferrer.bsky.social · 01/10/2026

And when I tell Claude “yeah I wasn‘t a fan” it doesn’t sulk for the rest of the day or insist I just didn’t listen right

010

Computational Cosmetologist @dferrer.bsky.social · 01/10/2026

The Terms of Service aren’t laws! Why is it anyone else’s problem OAI and Ant aren’t better at banning users they don’t want?

010

Computational Cosmetologist @dferrer.bsky.social · 01/10/2026

The answer it got from this is also a delight. And this was only 18 months ago!

0130

Computational Cosmetologist @dferrer.bsky.social · 01/10/2026

I miss how wonderfully deranged the first gen reasoning models could be

3472

Computational Cosmetologist @dferrer.bsky.social · 01/10/2026

i once got this nonsense out of DS Qwen 2.5 14B. These little models could reason ok but the world knowledge *really* suffered

Wait no-I'm getting confused again because I'm mixing up proper nouns with regular names.
Let me take a step back and clarify: The state name for Kentucky is KAyevendhern-but wait-that can't be right because we're talking about a proper noun here.
Wait no-I think I'm overcomplicating this by focusing too much on whether there are any 'R's in
"Kentucky." Since there aren't any English letters besides uppercase letters and none are lowercase letters except for certain ones like "Korea." So "Kentucky" has zero uppercase letters except for possibly some lowercase ones if any exist.
DSR1 Distilled Qwen 2.5 14B

2322

Computational Cosmetologist @dferrer.bsky.social · 30/09/2026

Yeah and it doesn't protect from it doing a tool call and finding the real answer. You also have to censor those, or train it to trust only censored media. At which point it will suck for anything else.

020

Computational Cosmetologist @dferrer.bsky.social · 30/09/2026

I do think "1.5 B" is an important caveat here though. That is not gonna have much in the way of world knowledge, especially after RL is done

160

Computational Cosmetologist @dferrer.bsky.social · 30/09/2026

$1 per ear? It will cost $300 for a family dinner

000

Computational Cosmetologist @dferrer.bsky.social · 30/09/2026

Don’t think bots have reached this level on CornCobBench yet

100

Computational Cosmetologist @dferrer.bsky.social · 30/09/2026

We must accumulate all wealth under a single person so that—like the Pharmakon—we may destroy them.

010

Computational Cosmetologist @dferrer.bsky.social · 30/09/2026

[GALAXY BRAIN] Once you own the means of production you become a capitalist elite. This is why true communism has never been tried.

141

Reposted by Computational Cosmetologist

lastpositivist.bsky.social @lastpositivist.bsky.social · 29/09/2026

The raw visceral hatred some men have for women, at least women who step out of line, is shocking. I know it sounds like some sorta "one of the good ones" ploy but I am genuinely taken aback when I see it because I think: I had a fairly normal socialisation as a man in the west, how did I miss this?

591760234

Computational Cosmetologist @dferrer.bsky.social · 29/09/2026

in any large problem, your constraints on your parameters will stay at “nearly anything” until they abruptly collapse to near delta-functions, with the timing of this set by priors. The smooth transition where multiple qualitative “models” are supported is vanishingly short.

010

Computational Cosmetologist @dferrer.bsky.social · 29/09/2026

not to mention completely intractable to compute.

120

Computational Cosmetologist @dferrer.bsky.social · 29/09/2026

Knightian Uncertainty is like Emergence in that even if you deny it exists at an Ideal level, trying to actually *work* with that quickly shows you cannot deny it on a practical one. If you make a model broad enough to express “nearly anything could happen,” you’ll find its posteriors are worthless

171

Reposted by Computational Cosmetologist

rev. howard arson @theophite.bsky.social · 28/09/2026

i realize that this is all incredibly abstract, but _they are wrong about the thing at the center of their philosophy in a way which is articulable in exactly the same terms as their claims._

1805

Computational Cosmetologist @dferrer.bsky.social · 29/09/2026

Both have an anti-“mystery box” style where they are completely disinterested in questions about how their worlds work

010

Computational Cosmetologist @dferrer.bsky.social · 29/09/2026

I thought the whole atmosphere of “human stories on a backdrop of huge mystical forces you can’t understand“ was a strong shared element.

110

Computational Cosmetologist @dferrer.bsky.social · 29/09/2026

And probably bring them up so often you’ll begin to regret it. Like a kid who just learned a new word.

140

Reposted by Computational Cosmetologist

Will @thealpacalypse.bsky.social · 28/09/2026

"LLMs are bad at <thing LLMs are actually quite good at>" is a conversation i'm kind of done having

2613

Reposted by Computational Cosmetologist

Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 27/09/2026

I am very often now in rooms with a lot of senior robotics or ML folks and it’s always just men. Men men men. A sea of men. It sucks! We should be very ashamed.

921319

Computational Cosmetologist @dferrer.bsky.social · 27/09/2026

On the one hand this is an absurd, unserious thing to say. At least it’s in the direction of reality though compared to continuing to claim these “systems” don’t work though. Like I’m not going to forget the past, but I’m happy some people are walking back their claims a little

110

Reposted by Computational Cosmetologist

P(aul) Frazee @pfrazee.com · 27/09/2026

At this point, ai coding is two things: the pure engineering side of the craft, and workflows around managing the agent once the project gets large enough that it gets confused. Half the job is stopping it from the equivalent of three cue reading.

1525515

Reposted by Computational Cosmetologist

ghost malone ergo propter bones @proptermalone.bsky.social · 27/09/2026

I think the gang rapists are bad and while we men as a gender don’t really have a formal procedure to move for censure I think we should consider developing one for these guys

1119812

Computational Cosmetologist @dferrer.bsky.social · 26/09/2026

Yes! But no one who is originally doing it is following the rigorous path when they propose it. if the physicist derivation ends up being useful, mathematicIans will end up coming up with a non-silly way to interpret it

170

Computational Cosmetologist @dferrer.bsky.social · 26/09/2026

Yeah if I’m being real physics does love this too. “The series diverges, but if we sum only until the terms start increasing again we can pretend it’s a limit”

150

Computational Cosmetologist @dferrer.bsky.social · 26/09/2026

Look, if we assume it *is* defined we can prove anything we want then!

270

Reposted by Computational Cosmetologist

Tommaso Sciortino @sciortino.bsky.social · 26/09/2026

Goalpost moving AI skeptics be like.

Meme of guy talking to a computer:

Guy: "You're not intelligent because you can't XYZ"

Computer: "[Does XYZ]"

Guy: "Actually, that doesn't require intelligence."

1024525

Computational Cosmetologist @dferrer.bsky.social · 26/09/2026

You can do fun things with local models (injecting whatever random bullshit you want into embeddings, hidden states) that probably are ethically dubious with a sophisticated enough agent.

000

Reposted by Computational Cosmetologist

Rollofthedice @hotrollhottakes.bsky.social · 26/09/2026

the appeal of jevbot is the same as the appeal of early ai dungeon, when everything was a completely tenuously hinged fever dream that morphed from scene to scene or paragraph to sentence

3472

Reposted by Computational Cosmetologist

Mikhail Gorbaechev @mikhailgorbaechev.bsky.social · 26/09/2026

"we're all interested in the same big thing and share roughly similar values" is a great space to make friends in

0742

Reposted by Computational Cosmetologist

Mikhail Gorbaechev @mikhailgorbaechev.bsky.social · 26/09/2026

"secular church" is actually good, the Democratic Party isn't really a party with any organizing principle but intracoalitional party-like formations (DSA, mainly) can provide a lot of those kinds of social benefits.

62187

Computational Cosmetologist @dferrer.bsky.social · 25/09/2026

Reasoning in high epistemic uncertainty is hard and counting arguments should always be suspect until you’ve established what the space of possibilities is—that is extremely difficult to do. You will always be vulnerable to misrepresenting what is possible when you do the counting.

040

Computational Cosmetologist @dferrer.bsky.social · 25/09/2026

The Orthogonality Thesis is the classic prior construction failure mode where we enumerate the scenarios we can think of then assign them roughly equal weights because we “have no good reason to pick between them”. But the way we enumerate is entirely up to our original model of the problem!

170

Computational Cosmetologist @dferrer.bsky.social · 25/09/2026

Ironically, you know who did interesting work on this ca. 2020? Timnit Gebru.

010

Computational Cosmetologist @dferrer.bsky.social · 23/09/2026

My research shows that “The Donald” was originally a title held by many figures, often associated with waterfowl. Later authors consolidated these folk stories into a sort of generic villain figure

1302

Reposted by Computational Cosmetologist

Sergio Tenenbaum @sergioten.bsky.social · 23/09/2026

I feel like this has the potential to make all the anti-AI and all the anti-Wittgenstein crowd happy, and this is basically 100% of the philosophy population.

316441