Sabitlenmiş Tweet
lucy 🐧
8.5K posts
lucy 🐧
@uneventual
i’ll die with a hammer in my hand
sf Katılım Aralık 2020
1.6K Takip Edilen1.9K Takipçiler
@YafahEdelman one way i think about this is opus 4.5 is like a 446 elo chess player* but i suspect there’s no set of tokens you could put in its context to make it a 1500 elo chess player, but apparently this happens pretty regularly with moderately smart humans
*maxim-saplin.github.io/llm_chess/
English
I don't get why everyone is talking about continual learning so much. The original GPT-3 paper was all about and how models can learn in-context, and they've only gotten better since then.
English
@RokoMijic i think a mixed strategy is probably best here but kind of is ruled out by construction in newcomb’s?
English
i think the best restatement of the 2-boxer case is to imagine being a submarine crew that’s watched its country obliterated and now must choose whether to retaliate—whatever good your character as an agent might have done is in the past, and now you have a causal choice
Invisible Hand Fluffer@maxflowminclout
@mayaofspring @socialtranxiety from 2009 to 2020, it seems like decision theorists shifted significantly in favour of two boxing
English
@princess_worms @YosarianTwo @madeofmistak3 i think it’s reasonable to say something like, your homeland is totally obliterated and so it’s not iterated
English
English
@YosarianTwo iiuc it’s omniscient not omnipotent, i think applying a perfect predictor here would mean that it’s simply impossible for you to retaliate after your adversary predicts you won’t and nukes you, but that doesn’t make a lot of sense to me.
English
@uneventual @madeofmistak3 Well the difference is that in nuclear MAD the opponent isnt omnipotent and doesn't ever know if you'll retaliate or not, right? Newcomb's is special because it's an omnipotent opponent.
English
wait i think it’s not hard to tell ai from human here (i went 4/5 without trying) and this is just evidence americans are piggies who love slop. theyve been doing these studies since the gpt-3 era and preference for slop has been a consistent finding.
Kevin Roose@kevinroose
judging from responses to the AI vs. human writing quiz, twitter appears to be in the bargaining/depression stage of the kubler-ross process, while bluesky is firmly in the denial/anger stage.
English
new sign of the endtimes: yr status page looks like a fucking wordle
SPEC@___4o____
OAI and Claude both dropped to 98% uptime during February. Another data point: Github has had more outages in Q1 2026 than the entirety of 2016-2019, according to their status page. Software is objectively getting worse.
English
Occasional reminder: p(doom) is an anti-helpful concept because it conflates the danger of AI with the probability that rush into it. And the latter is weird to estimate: facts of computer science that govern AI are fixed, but whether we stop is still up to us.
English