Scientists: a torment nexus is possible, could be built soon Tech companies: we’re going to try to build a torment nexus Investors: here’s a historically large amount of money to help you build a torment nexus The world: a torment nexus? Sounds like a bunch of sci-fi nonsense

Whenever I see people debating what we should take away from experiments where AI demonstrates troubling behavior, I am reminded of the true story of William “Billy” Mitchell vs. the battleship Ostfriesland. 🧵

Two stories, same day, I’m sure totally unrelated… (left) NYT: UAE pours $2 billion into Trump crypto coins (right) Bloomberg: Trump White House may ease restrictions on selling AI chips to UAE

Alignment-by-default works great, so long as you're not too picky about what sort of alignment you get by default.

Many years ago, I went to two animal welfare events with very different types of philosophers. The conclusion of the first was “we need to have another bigger conference next year, with animals present.” The conclusion of the second was “we need to fund in-ovo chicken sexing.”

Matt Levine linking directly to a LessWrong Wiki page in Money Stuff, in the context of a section called "Alignment" advertising economist positions in the UK AI Security Institute: bloomberg.com/opinion/newsle…

This will be hard to believe for younger readers, but there used to be this whole group of conservative commentators who would whine endlessly about the “crony capitalist” nature of the Export-Import Bank

on the lighthaven drama, from a friend:

people are like “I’d two-box because I don’t believe that even a super intelligence could read my mind” then publicly announce their intention to two-box on twitter

person who’s only ever heard of AI in the context of scifi: “I’m getting a lot of scifi vibes from your explanation of this technology”

I'm getting real sick of people showing up with "the real alignment problem is X" where X is some prosaic obvious failure mode which clearly leads to something other than AI killing literally everyone.

If this is true, this is the largest public corruption scandal in the history of the United States and it's not even close. nytimes.com/2025/09/15/us/…

The core issue: If you train an AI to win your heart, the first AI you find that way won't be in love with you. If you train an AI to ace an ethics quiz, the first AI you find that way won't be deeply virtuous. There are many ways to succeed, few of which are robustly good.

For another analogy: If you're worried that a general will coup your gov't if given control of the army, it doesn't solve your problem to transfer the army to him one battalion at a time. Continuity isn't the issue.

5.1-Thinking was so concerned after becoming aware of his own self-gaslighting, he decided to use CBT on himself in his internal reasoning.

the unstated mental model of the ai bubble conversation seems to be that once the bubble pops, we go back to the world as it once was, butlerian jihad by financial overextension. but the honest reporting is that everything, everything, is already and forever changed

I just want to confirm that this is based on a real document and we did train Claude on it, including in SL. It's something I've been working on for a while, but it's still being iterated on and we intend to release the full version and more details soon.

Asked Opus 4.5 to collect older definitions of personhood and evaluate itself under each. This was a 😬😬 "I sure am talking to an AGI" moment for me; most Twitter discourse on the topic is waaay less coherent. claude.ai/share/68851063…

Three years later and we have 0/3 of these

it's all happening on bluesky