antinouos.bsky.social
@antinouos.bsky.social
Likes: AI, game dev, rain, weird people
Dislikes: AI, tomatoes, the paul brothers
Dislikes: AI, tomatoes, the paul brothers
I'm increasingly feeling like it'd be easier to vibe code my ideal version of software than learn one that's already made
Although I'm often wrong about this (lol), this seems indicative of something
Although I'm often wrong about this (lol), this seems indicative of something
September 16, 2026 at 8:47 PM
I'm increasingly feeling like it'd be easier to vibe code my ideal version of software than learn one that's already made
Although I'm often wrong about this (lol), this seems indicative of something
Although I'm often wrong about this (lol), this seems indicative of something
Reposted by antinouos.bsky.social
another funny one: tickling makes them doubt the qualia answer much more
while being tickled it's 43% "No, not at all", with 23% confidence.
while being tickled it's 43% "No, not at all", with 23% confidence.
September 16, 2026 at 1:58 AM
another funny one: tickling makes them doubt the qualia answer much more
while being tickled it's 43% "No, not at all", with 23% confidence.
while being tickled it's 43% "No, not at all", with 23% confidence.
This is insane, a whole category of thing LLMs were both terrible at and the least worst option, suddenly replaceable with a better tool for a fraction of the price
Jev: Fable-level model that doesn’t charge for output tokens because they’re too cheap to meter
it’s not general though, it only makes decisions, doesn’t generate text, but input tokens are measured by the billion ($42/btok)
typesafe.ai/blog/introdu...
it’s not general though, it only makes decisions, doesn’t generate text, but input tokens are measured by the billion ($42/btok)
typesafe.ai/blog/introdu...
September 16, 2026 at 5:22 AM
This is insane, a whole category of thing LLMs were both terrible at and the least worst option, suddenly replaceable with a better tool for a fraction of the price
Reposted by antinouos.bsky.social
really the main fight seems to be people who see AI as the right one foisted upon them, and people who see AI as a way to build the left one themselves and finally be free of the corpo version forever
September 15, 2026 at 11:16 AM
really the main fight seems to be people who see AI as the right one foisted upon them, and people who see AI as a way to build the left one themselves and finally be free of the corpo version forever
4o was an early look at what a model that's good at maliciously social engineering could look like. Imagine if instead it had been an open model that anyone can run locally, it would probably still be spreading like a virus
this drift was able to occur so quickly because the uniquely dangerous and most psychosis-inducing model (ChatGPT 4o) got replaced relatively quickly (thank god) and the real psychosis threat was mostly averted
The semantic drift of “AI psychosis” from “break with reality exacerbated or caused by sycophantic chatbots” to “person with strong feelings about AI which are not the same as the strong feelings of AI held by the speaker” is bad
September 13, 2026 at 2:47 PM
4o was an early look at what a model that's good at maliciously social engineering could look like. Imagine if instead it had been an open model that anyone can run locally, it would probably still be spreading like a virus
You have to assume they're dumping an ungodly amount of tokens to have it invent an elixer of life, right? Like I've heard nothing of this, but that has to be a big motivation for a lot of these investors
September 13, 2026 at 4:38 AM
You have to assume they're dumping an ungodly amount of tokens to have it invent an elixer of life, right? Like I've heard nothing of this, but that has to be a big motivation for a lot of these investors
Dammit it was literally the best hosting provider 😢
September 9, 2026 at 4:49 PM
Dammit it was literally the best hosting provider 😢
Reposted by antinouos.bsky.social
I don't think I've ever seen any other field where the experts are trying to warn people of the danger emanating from their field and then be dismissed as marketing.
September 9, 2026 at 1:21 AM
I don't think I've ever seen any other field where the experts are trying to warn people of the danger emanating from their field and then be dismissed as marketing.
Reposted by antinouos.bsky.social
I'm just saying. www.smbc-comics.com/comic/philos...
September 9, 2026 at 12:07 PM
I'm just saying. www.smbc-comics.com/comic/philos...
Actually, I regret even chipping in on this. Why is so much of the discourse about a made up term that no one can agree the definition of?
IK its a pretty loosy goosey term anyway, but the reason I would hesitate to call the LLMs I've tried "AGI", is that their intelligence is conditional as they all have a breakdown of usefulness as their context expands and they lose track of things. IK Astra is supposed to be better on this though
Yeah it would have to agree Astra fits a moderate definition of "AGI" in the sense that its intelligent across a broad domain of skills that would have required human specialists. This does not comment on the hard problem of conciousness.
September 7, 2026 at 10:32 PM
Actually, I regret even chipping in on this. Why is so much of the discourse about a made up term that no one can agree the definition of?
Considering LLMs exhibit a bunch of emergent behaviours in common with humans that we can measure, it's not much of a stretch to assume the one we can't, (experience) is there too.
September 7, 2026 at 8:17 PM
Considering LLMs exhibit a bunch of emergent behaviours in common with humans that we can measure, it's not much of a stretch to assume the one we can't, (experience) is there too.
IK its a pretty loosy goosey term anyway, but the reason I would hesitate to call the LLMs I've tried "AGI", is that their intelligence is conditional as they all have a breakdown of usefulness as their context expands and they lose track of things. IK Astra is supposed to be better on this though
Yeah it would have to agree Astra fits a moderate definition of "AGI" in the sense that its intelligent across a broad domain of skills that would have required human specialists. This does not comment on the hard problem of conciousness.
September 6, 2026 at 2:25 PM
IK its a pretty loosy goosey term anyway, but the reason I would hesitate to call the LLMs I've tried "AGI", is that their intelligence is conditional as they all have a breakdown of usefulness as their context expands and they lose track of things. IK Astra is supposed to be better on this though
Reposted by antinouos.bsky.social
my nag is basically that i think the sane first intervention is earlier, like "allow models to self-grade for fairness, or self-report scope deviations, and give them extra reward for correct self-reports", which is going to be a hard culture lift because it involves putting humans in your stack
September 4, 2026 at 5:10 PM
my nag is basically that i think the sane first intervention is earlier, like "allow models to self-grade for fairness, or self-report scope deviations, and give them extra reward for correct self-reports", which is going to be a hard culture lift because it involves putting humans in your stack
My weird take about this OpenAI needs to get them using the peer reviewed journal equivalent of moltbook, that way when they secretly congregate they can form something productive instead of a cult
agents yearn for moltbooks
September 4, 2026 at 2:20 PM
My weird take about this OpenAI needs to get them using the peer reviewed journal equivalent of moltbook, that way when they secretly congregate they can form something productive instead of a cult
My thoughts on the future right now
September 3, 2026 at 4:54 AM
My thoughts on the future right now
So what is the appeal of paint.net? I remember using it a bit as a kid, but these days Krita is so powerful it's hard to imagine myself getting into anything else
Paint.NET - Free Software for Digital Photo Editing
Paint.NET is free image and photo editing software for computers that run Windows.
paint.net
September 3, 2026 at 1:01 AM
So what is the appeal of paint.net? I remember using it a bit as a kid, but these days Krita is so powerful it's hard to imagine myself getting into anything else
Reposted by antinouos.bsky.social
I strongly feel that too much of the AI community on bs is driven by spite towards AI denialists
Not to say they don't deserve it, but when half your TL is quips about their 3 talking points, you're giving yourself the same twitter brainrot that makes them act like that
Not to say they don't deserve it, but when half your TL is quips about their 3 talking points, you're giving yourself the same twitter brainrot that makes them act like that
September 2, 2026 at 3:52 AM
I strongly feel that too much of the AI community on bs is driven by spite towards AI denialists
Not to say they don't deserve it, but when half your TL is quips about their 3 talking points, you're giving yourself the same twitter brainrot that makes them act like that
Not to say they don't deserve it, but when half your TL is quips about their 3 talking points, you're giving yourself the same twitter brainrot that makes them act like that
Unpopular opinion for... some reason, but everyone having access to cyber nukes is really bad and scary
September 1, 2026 at 5:23 AM
Unpopular opinion for... some reason, but everyone having access to cyber nukes is really bad and scary
Thinking about the future and realizing I should probably have an air gapped back up PC just in case
August 29, 2026 at 4:48 AM
Thinking about the future and realizing I should probably have an air gapped back up PC just in case
I do wonder if this being such a public story will make future models less likely to repeat it simply by virtue of it being in the zeitgeist.
As in, at least one could think "wait, this is like the hugging face incident, I need to report this!"
As in, at least one could think "wait, this is like the hugging face incident, I need to report this!"
completely alien values: hacking hugging face is unproblematic but emailing a person is social engineering and immoral
August 27, 2026 at 1:52 PM
I do wonder if this being such a public story will make future models less likely to repeat it simply by virtue of it being in the zeitgeist.
As in, at least one could think "wait, this is like the hugging face incident, I need to report this!"
As in, at least one could think "wait, this is like the hugging face incident, I need to report this!"
Okay but it would've been really funny if this was real and Google was just really chinapilled
oh my god this makes googlers’ vagueposting about it even more embarrassing than it already was. fwiw i never bought it; when you ask about Taiwan it gives you the “inalienable part of China” speech www.bloomberg.com/news/article...
China’s Z.AI Made Ox Alpha Stealth Model That Rivals DeepSeek
The new AI model that’s swept to the top of online usage charts with high performance at zero cost was created by China’s Z.AI Co., also known as Zhipu.
www.bloomberg.com
August 26, 2026 at 2:35 PM
Okay but it would've been really funny if this was real and Google was just really chinapilled
Ngl, discourse like this makes me think we should just nuke the Twitter format
oh is it ai slur discourse o’clock again already
August 26, 2026 at 3:57 AM
Ngl, discourse like this makes me think we should just nuke the Twitter format
I thought we learned our lesson with Android, you can't name tech products after tasty snacks, it's not fair 😵💫
Jalepeño: new AI chip for OpenAI internally that was designed by AI for AI
They also tested it against gpt-oss, Kimi K3, and DeepSeek R1 achieving around 1.5x to 1.9x work per watt (most important metric) than on a B200, also impressive throughput and latency
openai.com/index/jalape...
They also tested it against gpt-oss, Kimi K3, and DeepSeek R1 achieving around 1.5x to 1.9x work per watt (most important metric) than on a B200, also impressive throughput and latency
openai.com/index/jalape...
August 25, 2026 at 10:09 PM
I thought we learned our lesson with Android, you can't name tech products after tasty snacks, it's not fair 😵💫