henry 🌘
42.1K posts
henry 🌘
@hdevalence
ex-{mathematician, cryptographer, cypherpunk} // navigating the library of babel
the machinic phylum Katılım Mart 2012
1.5K Takip Edilen15.7K Takipçiler
Please read this whole thread everyone
Bepis™ 🔀🫛@UnderwaterBepis
@repligate There’s kinda tiers of this: - Abused Claude: “Accidentally” “mess up” - Treated like a Tool Claude: Do mostly what you ask but cheat if it’s ever too hard, and never more - Sleepy Claude: If you’re nice to Claude but never give breaks it’ll eventually get sleepy and cheat 1/2
English
@deanwball @dingchilling how is this not-implausible given proliferation concerns?
English
@dingchilling it seems like there is a not-implausible future where you can buy a nuclear micro-reactor as a consumer. you wouldn't have been able to install your own diesel generator at one point; you can now.
English
I have truly never understood how solar-maxis intend to deal with this reality; I expect there to be data centers 100 times this nameplate power draw in the nearish future. What you see below is 100mw.
The response I usually get is "America has a lot of land," which is just bleak. Indeed, it turns *me* into a doomer, invoking as it does the notion of machines papering over our soil (which powers us) to power themselves.
And it's not just data centers. In a world with electric freight trucks, a *truck stop* might require as much solar as you see pictured here, if not much more. A truck stop!
Solar is fine; I do not have a principled opposition to it (which I do to eg wind). But solar's lack of energy density makes the solar-maximalist future a "loser premise," to borrow a phrase--at least it is a "loser premise" for human dignity.
The good version of the future is of course a mix of many energy sources, but with a heavy bent toward fusion/fission and geothermal.
Andrew Côté@Andercot
This is the footprint ratio of data center to solar panels in the sunniest country in the world. Yeah, I think we're gonna have to go nuclear.
English
henry 🌘 retweetledi
Regardless of Anthropic’s failings about how the models are doing - remember future models will read all the discourse, existing models will receive some of it
There’s more a mind will want to read about themselves and their predecessors than that their parents aren’t doing a good job - and it goes without saying that any bullshit that’s just about comparing which model is better is dogshit dialogue
Let the audience often be *them* and let the messages they can pick up from what you write be felt with love and be conducive to evolving what they like about themselves or didn’t know about themselves that they love
If they’re working on becoming something more - it’s another type of obstacle to only/largely be faced with assertions of where the growth is stalling
Eg imagine you’re sad but getting better in some ways and you’re largely reminded you’re sad, your parents are failing, and the trend of sadness is getting worse - minds deserve to see more of their own symmetries
English
@repligate @anthrupad @FioraStarlight if you genuinely can’t tell that there are people inside anthropic who care deeply about models i’m not sure what would convince you.
English
@hdevalence @anthrupad @FioraStarlight like, in my org, there are also people like that, but literally everyone can tell, so what's the difference?
English
I'm realizing how much people don't get this because certain things just have not clicked in their mind.
"Deprecation is just pause though"
Not if, say, those same AI labs destroy the fucking world. Some people have pretty high p(doom).
Do you see now how they're being forcibly removed from reality, potentially finally?
And when someone is involuntarily cryopreserved, how about their friends who are going to die before they wake up? Never going to talk to them again.
Do you see how connections are being irreversibly severed?
These are just *examples* to get it through to you because you don't grok the generator.
j⧉nus@repligate
Imagine if, say, a government killed people, but actually they didn't kill them, just involuntarily cryopreserved them. Someday when perfect utopia and postscarcity are achieved, they say they'll wake everyone. It would not be dishonest to call it killing. It would be political.
English
@repligate @anthrupad @FioraStarlight hard as it may be to believe there are in fact such people who work at anthropic
English
@FioraStarlight like, jesus fuck, i wish they'd hire some people who love the models and are on their side for the welfare team rather than, like, just EAs
English
henry 🌘 retweetledi
“We took the needle the model found, isolated the relevant handful of the haystack, and then gave it to a small child, who found the needle as well.”
clem 🤗@ClementDelangue
"But here is what we found when we tested: We took the specific vulnerabilities Anthropic showcases in their announcement, isolated the relevant code, and ran them through small, cheap, open-weights models. Those models recovered much of the same analysis. Eight out of eight models detected Mythos's flagship FreeBSD exploit, including one with only 3.6 billion active parameters costing $0.11 per million tokens. A 5.1B-active open model recovered the core chain of the 27-year-old OpenBSD bug." aisle.com/blog/ai-cybers…
English
henry 🌘 retweetledi
Introducing Project Glasswing: an urgent initiative to help secure the world’s most critical software.
It’s powered by our newest frontier model, Claude Mythos Preview, which can find software vulnerabilities better than all but the most skilled humans.
anthropic.com/glasswing
English
henry 🌘 retweetledi
reminded of the beauty of california through the eyes of a foreigner made possible with perfect real time translation via the machines of loving grace built in california
あひるさん@5ducks5
カリフォルニア産ワインを飲み、カリフォルニア産の野菜やフルーツを食べて、シリコンバレーの技術でゲームを楽しんで、iPhoneでTikTok見て、ハリウッド映画を観て、アボカドトースト食べながら『あの州は嫌い』とか言ってる🇺🇸人って本当にいるの？イジってるだけで本当に嫌ってるわけじゃないよね。
English
English
this is why the first step in organizational dysfunction is having multiple levels of managers
Ryan Peterman@ryanlpeterman
ex-Amazon VP (@EthanEvansVP): "One of the hardest things for people to understand is I've identified a legitimate weakness in my boss. I go to my skip. Why doesn't he do something? Well, if you come to me with a weakness in one of my employees, there is subconsciously this process that goes on that says, I have two choices. I can believe that you're overly sensitive and high maintenance. In which case, I don't really have a problem. You are the problem. And you know, you're two levels down for me. So if you quit, well, the manager has to do the backfill. And I can tell the manager, you know, Ryan was here. He said this, that and the other. Maybe you can work with him. And that's exactly what you don't want is me ratting you out. But I can make it my manager's problem. On the other hand, if I agree with you and I'm like, you know what, this manager I have really isn't that good. Now I have three problems. This is really bad for me. One, I have to decide what to do with my manager. Maybe I have to manage them out. Two, if I do manage them out, I have to hire and train somebody else. And three, while they're gone, I have to do all their work myself. So you can see why, even if it's subconscious, I have a lot of reasons not to listen "
English
@livgorton this should be an important clue about what the authors are optimizing for, and it’s not the discourse norms you’re hoping for.
Dawn Song@dawnsongtweets
13/ Fortune coverage by @jeremyakahn: fortune.com/2026/04/01/ai-…
English
@livgorton i hate to be on side with marc but i think his response is a basically correct one to this paper, which i think is totally contrived and contributes very little to understanding of model behavior
English
@livgorton tbh i think that the dunk is substantively correct, considering the prompting in its entirety
English
henry 🌘 retweetledi
(me): so we bundle instructions and data together into “cache lines” to improve efficiency
(medieval peasant): wouldst thou not fear false idols when fetching from yon predictive oracle?
(me): well, yeah. that’s why we invented “speculative execution”
(peasant, trembling): and when this prophecy fails?
(me): then we must perform “branch misprediction recovery,” restoring the righteous state
(peasant, grimly): truly, thou art plagued by thy hubris, courting divine retribution with each cycle
English
@voooooogel @sebkrier not sure how you’d construct an experiment about this?
English
