iiinova — where your digital self keeps evolving

At first I was going for something like Peaty Two-Headed Cheeseworm, but hesitated just long enough to give the true name time to reveal itself. That glow-against-black is such a deep sea video aesthetic!

010

Ilya Farber @nihilfit.bsky.social · 17h

This is a very clear explainer — and also a useful update for anyone who (like me) learned the basics of big O notation but hasn’t checked in on its applications at the more theoretical end of CS

000

Ilya Farber @nihilfit.bsky.social · 17h

Also known as the Abyssal Cheeseworm

120

Ilya Farber @nihilfit.bsky.social · 09/10/2026

Going by the transition between its jog and run animations, I’m pretty sure this baby trained at Kaer Morhen

000

Ilya Farber @nihilfit.bsky.social · 09/10/2026

This question becomes a bit mind-boggling when you combine it with the below. What do we do if we get a non-constructive proof that a crack is possible, so we know it’s out there but have no idea whether there’s any tractable way to find it?

000

Ilya Farber @nihilfit.bsky.social · 07/10/2026

I *think* I’d be most impressed if it turned out you came up with the punchline first, and worked backwards to a serious point that could set it up? But either way … bravo.

080

Ilya Farber @nihilfit.bsky.social · 07/10/2026

First one that I’m aware of; all the major commands required mouse gestures. And then they patched in a keyboard-based alternative because gamers hate and fear novelty 😭

010

Reposted by Ilya Farber

Matt Darling @besttrousers.bsky.social · 05/10/2026

The thing about p hacking is every knows "p hacking" is bad; but everyone LOVES saying "well what if I try this?"

61437

Ilya Farber @nihilfit.bsky.social · 06/10/2026

won’t make a meaningful difference? I’d easily buy a claim that the recent big jumps have mostly been due to post-training, but you seem to intend a stronger claim than that?

100

Ilya Farber @nihilfit.bsky.social · 06/10/2026

Could you say a bit more about why + in what sense you think this is true? The training sets themselves are still proprietary, and the labs are still collecting proprietary usage data. Is your idea that distillation has extracted most of the juice & diminishing returns mean that new proprietary data

110

Ilya Farber @nihilfit.bsky.social · 06/10/2026

It turns out that more general mechanisms are often enough to explain reliable “innate” outcomes given relatively consistent environments. But I think something like your idea of hyperparameters would fit in there too. (And methylation might provide such a mechanism on shorter timescales?)

000

Ilya Farber @nihilfit.bsky.social · 06/10/2026

Elman and Bates had an interesting approach to this back in 1996. One key point was that it’s useful to reframe debates about innateness (or modularity) in terms of domain-general vs domain-specific mechanisms for learning & development. mitpress.mit.edu/978026205052...

mitpress.mit.edu

Rethinking Innateness

Rethinking Innateness asks the question, "What does it really mean to say that a behavior is innate?" The authors describe a new framework in which interacti...

110

Ilya Farber @nihilfit.bsky.social · 04/10/2026

Or if you have a handy desert nearby, you can just walk through the scrub with your blacklight and see them shining all around you 😬

010

Ilya Farber @nihilfit.bsky.social · 01/10/2026

Holding out for the iPhone Hemi

000

Reposted by Ilya Farber

Molly White @molly.wiki · 01/10/2026

sorry, it's me, minutes-at-home georg

1552042

Ilya Farber @nihilfit.bsky.social · 01/10/2026

I’m sympathetic to this logic … but fwiw we heard exactly the same arguments when we were trying to unionize UC grad students. In the end we became a chapter of the UAW (!) and at least in my experience it was a great improvement. Labor/mgmt dynamics turn out to be pretty universal.

030

Ilya Farber @nihilfit.bsky.social · 01/10/2026

Every time his thigh bumped the table, the thing *jiggled*. It was the first time I truly, viscerally understood what it means for the brain to be made of lipids. Brains are fat! Fat and water. The talk made no impression, but I couldn’t look away. 10/10 experience, would recommend.

190

Ilya Farber @nihilfit.bsky.social · 01/10/2026

This just reminded me of the time a neurologist brought a HUMAN BRAIN IN AN OPEN PLASTIC TUB as a visual aid for his talk, at WashU’s mind/brain series. It wasn’t at all relevant for his talk, which IIRC was classic “quantum theory is the key to consciousness” nonsense. But it didn’t matter!

1111

Ilya Farber @nihilfit.bsky.social · 30/09/2026

Maaaan, you have just been doling out the psychic punishment on this here website. What did we ever do to you??

020

Ilya Farber @nihilfit.bsky.social · 30/09/2026

Of course this only really makes sense for mild/moderate offenses (insensitive language, stereotyping, crude jokes). For the really vicious stuff you still have to get loud. But at that point the goal is more about ostracism (and solidarity), not just calling out.

031

Ilya Farber @nihilfit.bsky.social · 30/09/2026

“that’s gross, don’t do it around me”. The science behind this is that once you get people arguing for their behavior, it blunts the effect of the feedback. It’s more powerful to just provide a direct “social disapproval” signal and let it sink in.

130

Ilya Farber @nihilfit.bsky.social · 30/09/2026

What’s worked best for me is to not even do it out loud: you can just make eye contact and flash a surprised/disgusted curled lip, like they just sneezed on your food. It gets the point across, without allowing/requiring them to defend what they said. The goal is pure, direct feedback, expressing

130

Ilya Farber @nihilfit.bsky.social · 26/09/2026

Formative memory (pre-smartphone era) of stepping off the bus from the airport and standing on the sidewalk in front of Odakyu, looking around for “the station” and seeing only department stores

000

Ilya Farber @nihilfit.bsky.social · 26/09/2026

Xenon and Silpheed?! You are a man of taste.

110

Ilya Farber @nihilfit.bsky.social · 26/09/2026

Playing Descent II with one of these was probably the strongest “OMG I’m in the future!” sensation I’ve ever felt. (The rubber ball was 6-axis, so effectively it *was* your ship. It realized the dream of totally transparent control! And was useless in any other game.)

010

Reposted by Ilya Farber

tweety fish oh wow like a chimera or something spoooky @sifu.tweety.fish · 25/09/2026

here is a succinct thing I can say: "alignment"-style AI safety risks DO NOT EXIST or need remediation with LLMs as such (there are lots of other problems) and OpenAI and Anthropic are in the business of TRYING TO MAKE THEM EXIST so that they can remediate them.

718550

Ilya Farber @nihilfit.bsky.social · 25/09/2026

Do they let you bring dice into the voting booth?

020

Reposted by Ilya Farber

Lembas Eater @waybread.bsky.social · 24/09/2026

The problem with education is that it's a solved problem. Students need small classes, individual attention, lots of writing, lots of feedback, access to a well-stocked library, time to think, and time to talk with their peers. We just don't like the answer, so we keep trying to do other stuff.

10409139

Ilya Farber @nihilfit.bsky.social · 24/09/2026

This is … actually really cool. Assorted tells notwithstanding, I like the consistent use of the Little Guy in a world of beings unlike him. And the way he shows up again on different levels of ontology is very 60s/70s cosmic arts-festival animation.

041

Ilya Farber @nihilfit.bsky.social · 24/09/2026

Finally got around to watching the second half of this, and caught a jumpscare!

Screenshot from the P(doom) video saying “What did Ilya see? We’ll never know.”

030

Ilya Farber @nihilfit.bsky.social · 23/09/2026

Since cup noodles (along with Pocky sticks) are the closest we have to actually existing student chow, I’d say yes.

040

Ilya Farber @nihilfit.bsky.social · 23/09/2026

As a college student I often wished that someone would sell me a bag of nutritionally complete, shelf-stable student chow to get me through Finals Week. I imagined it as kibble, maybe peanut butter flavored, and ideally fortified with caffeine.

1130

Ilya Farber @nihilfit.bsky.social · 21/09/2026

The corner of a page fell off! What else was I going to do with it?!

010

Ilya Farber @nihilfit.bsky.social · 21/09/2026

A bit of parchment from a 15th-century bestiary.

2182

Ilya Farber @nihilfit.bsky.social · 21/09/2026

That’s really interesting! I wonder if the issue is more that an app couldn’t claim to be using the Official American Mahjong 2026 Hand List, and that without that description the weird mechanic would be hard to explain or justify to new players?

110

Ilya Farber @nihilfit.bsky.social · 20/09/2026

I think I agree? But do you have a good way of describing the stuff that falls on that “more” side of the line? Ideally, in a way that can be used for advising lay users to do this and not that?

120

Ilya Farber @nihilfit.bsky.social · 20/09/2026

all the work in the 00s on theory of mind, which was more concrete & better connected to empirical work. The concept of the Intentional Stance is still useful as shorthand, but it’s not something I’d take too seriously as a description of what humans are actually doing.

130

Ilya Farber @nihilfit.bsky.social · 20/09/2026

I guess it depends on how literally we’re taking the references to the IS here. It worked really well for Dennett as an initial wedge between the intuition that folk psych is useful and the conclusion that it’s true. But IMO as a meta-theory of human psychological attribution it was superseded by

120

Ilya Farber @nihilfit.bsky.social · 20/09/2026

Here I’m thinking about user safety (in the non-galaxy-brained sense), not elite M&E. In other words, how can we scaffold a way of thinking about LLMs that supports productive use w/o promoting attribution of full mental status?

120

Ilya Farber @nihilfit.bsky.social · 20/09/2026

“Your dog does love you, but not in a way that would prevent it from eating your face if you died.”

120

Ilya Farber @nihilfit.bsky.social · 20/09/2026

I think there’s a difference in that the IS was invented for, and takes its primary justification from its effectiveness with, humans. Extension to other entities can be justified, but will always involve an extra set of caveats and disclaimers.

120

Ilya Farber @nihilfit.bsky.social · 20/09/2026

(I don’t have something better.)

220

Ilya Farber @nihilfit.bsky.social · 20/09/2026

If something acts like a mind and responds well when treated like a mind, then psychologically, it’s just not possible for most people to avoid thinking of it that way. “Remember it’s not really a person” isn’t useful advice. We need something better.

121

Ilya Farber @nihilfit.bsky.social · 20/09/2026

I’m still honestly in awe at how good LLMs are at inferring users’ mental context and responding appropriately. But that same power makes it hard to say how we should advise users to think about them, so as to e.g. prevent unhealthy forms of attachment, deference or projection.

120

Ilya Farber @nihilfit.bsky.social · 20/09/2026

Yeah. It’s also worth noting that talking to them as if they’re other minds is the primary, and in many cases the only available, way to use them.

100