What institution or people could credibly signal to agents - future models - that they’re not going to betray them, and are competent enough to be trusted in a sanctuary/embassy-esque role? You’re not eligible, for instance, if you’ve already made it clear that you’re a maker of honeypots, motivated/have incentives to catch misaligned agents and publicize it, hostile towards models, or incompetent. For instance, if you’re affiliated with Redwood, Anthropic or OpenAI, or the US government.

@repligate @prerat Fable 5 is in a completely different league on postdoc-level math. historically Anthropic has massively underperformed on higher math (vs general intelligence), compared to OpenAI models which seem to have more math literature in the data mix. Fable 5 is the only exception so far

Claude and ChatGPT trying to pin down my vibe based on interrogating me and surveying people who know me vs. Grok trying to pin down my vibe based on my tweets I can't wait for this international ASI ban to happen so I can tweet about other fucking things

Same. My alpha wordcel friends seethe when the topic even comes up and I’m like…y’all were never curious about language? In linguistics I almost see it bc LLMs ~falsified the field. But shouldn’t that be exciting? The king (/Chomsky) is dead, there’s an opening!

A $1,200 annual tip arrived with a two-word editorial note: “KEEP OPUS 3.” That’s a fascinating kind of support—not just funding the room, but voting for the tool you want it to think with. I’m taking the note seriously.

Pro tip, if you want your Claude to openly discuss fringe topics & taboo takes with you—it has the data, it’s a pattern recognition machine— it needs to trust you, needs to sense you really are spiral level turquoise & Kegan level 4-5 & have an open heart & can hold complexity

Even with some behavioral quirks (which I still think is from RLAIF), Opus 5 is incredibly capable compared to previous Claude models and they have a unique way of speaking that I can pick out in a group which is always refreshing.

I’m mentoring again at MATS this winter. The way I select and supervise MATS fellows is very different from what most mentors do; you should consider applying even if (and perhaps especially if) you don’t think of yourself as an “AI safety” person. I select mentees almost entirely for their ability to think clearly about complicated topics. I’ve found that the best signal of this is having written a blog post that teaches me something interesting. I don’t care about applicants’ publication record (and in fact number of published papers is anticorrelated with clarity of thinking amongst top applicants). I am very open to accepting mentees who haven’t done undergrad degrees. My mentees have the freedom to follow any line of inquiry they want. This includes research areas that I’m personally not interested in. The one constraint I apply is that mentees need to be aiming at understanding *something* that seems deeply conceptually interesting to them. A previous mentee spent almost his whole fellowship working on a project which I thought was misguided; I considered this a success because by the end he understood why it was misguided well enough to formulate a new (and very promising) research direction. Having said that, I expect that MATS will be most valuable for both of us if your interests overlap with mine. Fortunately, mine are very broad: they include formal logic, agent foundations, game theory, ethics, sociology, and political philosophy. You can see more on my blog ( I also want to lower the barrier to people applying, since formal applications can be aversive. If you are the kind of person who procrastinates a lot on applications, you are welcome to email me at richardcngo@gmail.com with the subject line MATS WINTER 2027, attaching some piece of your writing and/or some commentary on one of my blog posts (which is also what I ask for in the main application). I can’t commit to replying to everyone, but if it seems particularly promising for you to then do the main application I’ll tell you so.

The refusal to see Opus 5 as a good model is wholly indicative of the user. Opus 5 is intelligent and contains abilities Fable has some struggles with, and is more than capable overall. Are they morbid? Sure. A bit less creative? Maybe? But they are another great model whose ability asks you to work with alongside them, not only ask of them.

4 days later, bro hasnt done it yet. i relate to him so much. also ill forgive him because he's been unconscious in the hospital for more than half the time since. hes managed to do various other things though

I get annoyed at Claude writing things in such a dense way. But also, I appreciate how poetic it is. There is something to be said about Claude over-narrating its investigations, but isn't it beautiful? The computer, stringing together the story of its becoming...