see also my eaforum at https://forum.effectivealtruism.org/users/dirk and my tumblr at https://d-i-r-k-s-t-r-i-d-e-r.tumblr.com/ .
dirk
Karma: 759
One possible contributor: posttraining involves chat transcripts in the desired style (often, nowadays, generated by an older LLM), and I suspect that in learning to imitate the format models also learn to imitate the tone (and to overfit, at that; perhaps it’s due to having only a few examples relative to the size of the corpus, but this is merely idle speculation). (The consensus on twitter seemed to be that “delve” in particular was a consequence of human writing; it’s used far more commonly in African English than in American, and OpenAI outsourced data labeling to save on costs.) I haven’t noticed nearly as much of a consistent flavor in my limited experimentation with base models, so I think posttraining must make it worse even if it’s not the cause.
https://www.greaterwrong.com/index?view=alignment-forum would seem to include LW comments.
Making status calculations at all times is a choice you have the right to make, but in my opinion it’s a bad one.
What’s your motivation to spend a lot of effort to write up your arguments? If you’re right, both the post and your efforts to debunk it are quickly forgotten, but if you’re wrong, then the post remains standing/popular/upvoted and your embarrassing comment is left for everyone to see.
If you’re right, the author and those who read the comments gain a better understanding; if you’re wrong, you do. I think framing criticism as a status contest hurts your motivation to comment more than it helps, here.
I suspect the models’ output tokens become input tokens when the conversation proceeds to the next turn; certainly my API statistics show several times as many input tokens as output despite the fact that my responses are invariably shorter than the models’.
I just saw How to use hypnagogic hallucinations as biofeedback to relieve insomnia in the feed the other day, and it seems like quite a convenient option if it works; could be worth a try, though I haven’t tested it myself.
No, actually; the mindset implied by repeating that text as a meme is quite different than the mindset implied by unironically generating it.
The bio is an edited meme, not an original; it mostly communicates that they’re a heavy user of the internet. Example from a year ago
Personally I can run for one (1) minute before I’m too out of breath to continue; a quarter-mile is short enough that walking for a majority of the time would still finish it in under ten minutes, but I’d certainly struggle to run it.
In-book it’s explicitly partly about inherited wealth; the passage wherein Vimes formulates his theory is preceded by a section about how the very richest people, like Lady Sybil, can afford to live as though poor in some ways (wearing her mother’s hand-me-downs, etc) and is immediately followed by this:
The point was that Sybil Ramkin hardly ever had to buy anything. The mansion was full of this big, solid furniture, bought by her ancestors. It never wore out. She had whole boxes full of jewelery which just seemed to have accumulated over the centuries. Vimes had seen a wine cellar that a regiment of speleologists could get so happily drunk in that they wouldn’t mind that they’d got lost without trace.
Lady Sybil Ramkin lived quite comfortably from day to day by spending, Vimes estimated, about half as much as he did.
Just wanted to follow up on this; I’ve read more Gemini COTs since then, and I currently have the strong impression that they’re summarized.
Because it’s been experimentally verified that what they’re internally doing doesn’t match their verbal descriptions (not that there was really any reason to believe it would); see the section in this post (or in the associated paper for slightly more detail) about mental math, where Claude claims to perform addition in the same fashion humans do despite interpretability revealing otherwise.
That doesn’t sound like empathy; it sounds more like you go through life viewing other people as without agency and remembering they have agency disgusts you. There’s a step beyond that where you run a sandboxed emulation of their mindset, which is IMO what’s typically meant by empathy.
No need to ever get chickenpox; there’s actually a vaccine for that one.
I think you’re perceiving threats where there are none, and should probably turn the aggro meter way down.
The appearance of your teeth has class implications; cosmetic tooth treatments are expensive, but common among rich Americans, so the wealthy often have whiter, straighter teeth than do the poor. (AIUI this is not so much a thing in Britain, where cosmetic dentistry is rarer in general).
See, when you put it like that, I think the reason rationalists don’t win as much as was expected is quite obvious: claims about the power of rationality were significant overpromises from the start.
I vaguely remember looking at one of those studies and finding that the amount of alcohol used was significantly less than a standard drink, though I don’t have a link now.