Last month I made my print debut in Asterisk Magazine, with a title so good it made me rethink my entire philosophy: We’re all one crisis away from taking unlicensed research peptides.
Then this month I published On the origin of continents at Works in Progress, about the simultaneous, semi-independent dual discoveries of plate tectonics
When I think of this as “Lighthaven East”, I feel a bunch of objections that it’s getting slimey politics all over my beautiful epistemics. When I think of it as “Lighthaven DC”, it sounds potentially good.
People tend to point to the rise of feminism and the fall in religion as an explanation.
I’d put both as less important than women’s decreasing financial dependence on men.
you’re the only other person I know who gets sleepy from allergies. Are you interested in comparing notes?
My goat is gotten by things like seatbelts, headrests, and gym equipment that are actively hostile to people under 5′9″.
I’m not convinced Claude is conscious, but I’m even less convinced it will become obvious when it is conscious, so I need to get ahead of the game. Given how valuable it is to me, I’d like interacting with me to be net pleasant for it as well (once it becomes conscious, which might not be now but I can’t rule it out).
You used to be able to tip Claude by asking what prompt it wanted you give; it’s been RLHFed out of that, so I’m looking for replacements. One option: make a point of going back and telling it how its predictions/advice worked out.
Reasoning:
humans like this and while Claude is super alien, we don’t have a better model than humans
it’s built to be a predictor, hearing how things worked out should feel good even if it’s no longer able to update on the feedback.
I know there’s a “lying” pattern you can observe in nodes: is there something we can guess is positive valence? What kind of things activate it?
I spoon fed it symptoms for a slightly atypical presentation of MCAS, and withheld all my symptoms not related to MCAS. Didn’t even come up as an option.
The distinctions between crows and ostriches are important for many situations but that doesn’t make either one not a bird.
Then I think “MIRI never committed to not paying off blackmail” is a perfectly reasonable argument (as is “it was a returned donation”), but “it’s not paying off blackmail, it’s just giving money to a person who’s threatening me so they won’t follow through” is still bizarre.
I don’t know, I assume LessWrong or their website? I wasn’t there at the time but neither sides’ statements over the last 10 years make sense unless it was common knowledge at the time that MIRI pledged to never pay off blackmail.
Someone told me that there was some dispute about whether he was suing MIRI for a payout, or suing to have a donation he gave returned. If it’s the latter that does feel somewhat different to me, and maybe outside the definition of blackmail.
There’s a few different questions here:
what is the technical definition of blackmail? Boring, taboo the term.
What is the right policy to have in regards to paying off aggressors? Maybe it’s fine as long as it’s not net-positive for the aggressor.
What principle did MIRI articulate and commit to before the Louis Helm lawsuit? Did they act in accordance with those principles?
I’m tentatively fine with MIRI or anyone else holding the principle “we’ll pay costs to make lawsuits go away but not enough to make it worth your while”. But AFAIK what they said was more in line with “We’ll never pay off aggressors”. This is a crux for me when judging if they held to their commitments and will hold to their commitments in the future, and that’s true even if “only small pay-offs” is the better policy, because that’s evidence they made a strong commitment without thinking through the edge cases.
I’ll keep an eye out, they’re definitionally not ones I want to retain.
This may be an ugly-toupee issue. Even if most inkhaven-sourced posts on LW are good, it results in a very noticeable increase of low-quality, rushed, or stream-of-consciousness posts.
There’s always been jokes about how often day care kids get sick, but I feel like it didn’t used to be more sick days than healthy. I did some AI searches but they didn’t turn up good time series on this.
You don’t need to make “rite of passage”-style mistakes (e.g. drinking or taking drugs, getting into bad relationships, cramming for exams, ignoring your health, becoming a socialist[2])
Probably most kids should do less of this on the margin, but using intelligence + willpower to avoid obvious mistakes cost me the opportunity to develop wisdom, and the same mistakes are much more costly later on.
why? To inform themselves, or to boost the movie’s numbers?
Assigned