Linus Mixson

Former gifted kid with ADHD, autism, low IQ, erectile dysfunction.

[Linus Mixson @LinusMixson]

@levon_whiteguy Reddit hyperbanned me a month ago because I was testing something work-related and it's been a huge, huge, monstrous relief.

[Linus Mixson @LinusMixson]

While I entirely agree what you've said in isolation, consider that "predicting" means something very different to different people. You or I might understand it as an operation defined against some hidden or partially-known function. Most people, however, understand it to mean something purely statistical, an in-distribution sampling against the corpus, something like a Markov model. It's unclear what Chiang understands it to mean, but what he says here seems to suggest the latter, and the difference between these two understandings is not at all immaterial.

[Linus Mixson @LinusMixson]

@teortaxesTex Taiwan and HK both have a huge boner for Japan, with a not-insignificant subset of Taiwanese people insisting that Taiwanese-Chinese are in some literal, if spiritual, fashion actually more related to the Japanese than Mainlanders.

[Linus Mixson @LinusMixson]

Really, really bizarrely unambitious concepts from Will here. Super strange.

[will depue @willdepue]

i really don't know what's going to happen to software in the future

some musings on a long flight:

i kinda just want to live in my beautiful custom software garden, touch no interface i don't control, just APIs and the agents i envoy to the comparably obscene unrestricted web. a digital immune system

i'm sure apps will just block this aggressively, unclear how successfully. there's some unlikely beautiful web2 world where everyone's individual ecosystem just receives the news/twitter/meta/etc. firehose and runs their own personal filtering & content system

it's unclear what social platforms survive in a fully custom/generative ui world. migration becomes a lot easier, i can both read + post across many without effort, and so can everyone else. obviously, there's a lot of format and culture to platforms, but these might get unbundled into protocols?

i just want:
1) most of the time, everything just comes through message + visual from my AI executive assistant. no other interface needed. voice input
2) just one messaging & mail app, can multiplex across platforms
3) every morning, a cross-platform social media brief
4) and if i want to go deeper, the appropriate social media specific interface but removed of ads, distractions, negative features
5) no need to reinvent the rest, i guess, but then again most apps suck. i just want good transcription in apple notes. spotify recs, music search, social could be so much better. i can't stand the new apple photos interface...

portability and collaboration are big issues for custom software, for sure, but these could be solved with more open protocols, and migration effort is vastly diminished anyways

there's some fear that, with a generative internet, the ai labs end up polluting the commons of god's chosen business (advertising) which supports the technology industry that they emerged from? maybe it doesn't matter at that point

ai video people occasionally talk about how you could at some point just live generate your entire computer with a video model. don't see why you would though really

[Linus Mixson @LinusMixson]

@packetsniffer99 @jagunanthi @joseph_h_garvin Right, the first point is the beginning and end of my case. The other is well beyond anything I'm saying.

[Linus Mixson @LinusMixson]

@thkostolansky This is going to take a bit of writing, so I'll try to get back to it after this meeting lol. To be clear, I'm a poor epistemologist, a worse logician, and a terrible engineer, so don't expect too much.

[Linus Mixson @LinusMixson]

I am not making a claim about action initiation, I am making the claim that the relationship of phenomenal consciousness to *any* observable is unfalsifiable. Action spontaneity is one of these observables. You're free to explain how it (falsifiably) demonstrates the presence or absence of consciousness, of course.

[Linus Mixson @LinusMixson]

@packetsniffer99 @joseph_h_garvin @jagunanthi Spontaneity has no falsifiable relationship to phenomenal consciousness and arguably has no connection to self-referential consciousness. Humans are, of course, also themselves not spontaneous in this sense: you didn't decide to be conceived.

[Linus Mixson @LinusMixson]

@packetsniffer99 @jagunanthi @joseph_h_garvin Irrelevant; no falsifiable relationship to consciousness.

[Linus Mixson @LinusMixson]

OpenAI needs to win because, when I jokingly asked an OAI engineer whether the models feel pain when you interrupt inference, he said "What? They don't feel anything." He's still alive today. I'm pretty sure that Anthropic throws you into the organelle separator if you do that.

[Linus Mixson @LinusMixson]

So you think that qualia are causal on physical reality? (Not an insane idea, to be clear, although it raises questions.) Or is it that you think there may just be overwhelmingly suggestive evidence for specific physical structures generating them, even if this connection can't ultimately be shown in a lab?

[Linus Mixson @LinusMixson]

Before I get into this, is your tactic here going to be to drill down on the semantics of any term until I say that something is merely intuitive or a matter of my own preference? Because I'll tell you straight up that, sure, under the gaze of eternity I'm doubtful my little preferences, what's interesting to me, is any more objective than anyone else's. At the end of the day none of it matters, right? But I'd challenge you to tell me that endless philosophizing about something that is in principle resistant to proof is a better decision than doing almost anything of practical consequence, under the notion of a "good decision" most of humanity holds.

[Linus Mixson @LinusMixson]

AI consciousness discourse is a way for people with no real technical ability to feel included in the phenomenon. Or merely stupidity.

[Linus Mixson @LinusMixson]

@thkostolansky As well as how to represent "in context" facts to a pure-reasoning system, of course.

[Linus Mixson @LinusMixson]

@thkostolansky Yes, as has been apparent from the earliest days of propositional logic. The hundred trillion dollar question is the factor by which crystalized knowledge of the world changes efficiency, and the factor by which it changes generality.

[Linus Mixson @LinusMixson]

If I'm correctly understanding what you're getting at (we should care about the existence of consciousness even if it's unobservable), I say — feel free. Enjoy the wallow in the realm of "maybe" while other people do something that's, for once, finally, after all this time, interesting.

[Linus Mixson @LinusMixson]

That's real king shit. Even DeepSeek hasn't done that in a year and a half.

[Linus Mixson @LinusMixson]

Holy shit, NVIDIA released the Nemotron base??

[wh @nrehiew_]

The base model basically crushes all other base models while admittedly the other base models are a bit old at this point

[Linus Mixson @LinusMixson]

@ImmaGott Yes, mine. Can you see my little qualia? Can you feel their effects? Do they impinge upon your world in any way?

[Linus Mixson @LinusMixson]

@packetsniffer99 @jagunanthi @joseph_h_garvin Unfalsifiable claim in the other direction.

[Linus Mixson @LinusMixson]

Because it's completely unfalsifiable, immaterial, insubstantial, and non-causal, and is the sort of familiar navel-gazing emotionally-appealing topic that only small minds would gravitate to when faced with a technology that has the potential to render life absolutely unrecognizable. It's like a bunch of hominids watching a guy snatch a blazing branch from a lightning-struck tree and then sitting around talking about "does fire… have feelings?"

[Linus Mixson @LinusMixson]

@jagunanthi @joseph_h_garvin I apologize for being so persistent in this. I've just seen a lot of self-important yapping and precious little attempt to answer "why," or even tolerance for the question, and I really appreciate that you're willing to engage on it.

[Linus Mixson @LinusMixson]

But what does it mean for an activation to behave like qualia? We already have activations that map quite neatly to sensory experience, but there's no observed or otherwise demonstrated causal pathway from those observations to the qualia themselves (not merely their presence or absence). And see my earlier point concerning these simplest system implementing these properties. Is a series of hydraulic valves responding to a cadmium photoresistor in a "qualia-like" way consciousness? Why or why not? Or is such a thing incoherent in terms? Then why?