ToValueFunfetti
ToValueFunfetti on Hacker News
The 'predictor' part referred to the fact that base models would effectively answer the question "what token is most likely next". Training away from that default behavior doesn't escape the technical meaning of those words- answering "what token is most likely next in a solution to Navier Stokes" is still predicting a next token- but at that point the label just means it outputs text serially.
I don't understand that what you mean. Disease and climate change seem like clear cut technology problems, and the rest, save for war, have obvious technological solutions. What does it mean for them to be human problems? Should I expect them to solve themselves?
I mean, ideally to the end of poverty, disease, climate change, war, starvation, etc. You can disagree about whether this would be the outcome in practice, but I don't think 'why might we want to create something smarter than us' is a hard question. Humans losing their status as top dog is really a trivial price to pay for the potential return.
I don't think the timelines make sense for it to have been included in training data, LLM training does not seem to be nearly efficient enough to learn these things from lone examples, and it wouldn't make sense for this earlier work to have made its way into both OpenAI and Gemini's data. Navier-Stokes is not strong evidence that this kind of thing can happen, as OpenAI denies that Buckmaster's work was in the training data for the model they used, Buckmaster didn't have a blowup that they could steal, and Scientific American reports[1] that the similarities between their approaches were superficial. It's also well-established that models can find novel vulns at this point.
[1]https://www.scientificamerican.com/article/openais-blockbust...
That may well be the case, in theory.
Anthropic has been going through the phase of doing this particular useless research project the entire time. The stated impetus for their founding is that OpenAI, which has also been doing this research the entire time, wasn't doing it enough. This shouldn't tell you anything about whether a bubble is about to pop.
The article is gone because the author retracted it
?
Surely, this is not correct. The monkeys are drastically less likely to produce valid words; the set of monkeys that produce valid words are much less likely to produce valid sentences, etc. etc. Very low bar to clear, but of course this is vastly better than infinite monkeys at producing good research.
How much did you get paid to write this?
Can you explain the need to be annoying about the YC 17 thing? What's the objective here? Is it supposed to convince potential founders to decide against taking YC's money as a way of putting financial pressure on YC not to fund projects that people don't like? Or is the intention that YC people are reading your comment directly and it's changing their minds?
I can't imagine either of these being impactful, and the effect seems to just be to distract from people's concerns about Flock (YC 17). Eg, my eyes are bouncing off of the presumably anti-Flock statement you're making here because you've coated it in repeated parentheticals and tortured the writing so you can say "Flock" as often as possible.
It's more like if the law says the maximum sentence for theft is 10 years, a thief appeals his 20 year sentence, wins, and gets out early. He goes and robs somebody else so you say the court was wrong to let him win.
The court's job is to uphold the law. If you disagree with their interpretation, you can call them incorrect. If you have a problem with the consequences of the law, you have a problem with the legislature.
I'm confused by this. Are you saying that everything related to the right wing and fox news is evidence that momentum is a good model or that trying to model history with momentum is fundamentally fallacious? If the former, you've really got to be more specific- was firing Tucker Carlson evidence for momentum somehow? Did Republicans conceding gay marriage or Trump conceding abortion give Democrats momentum?
Why am I being interpreted as advocating for abandoning the trans? I don't want to do that and I don't need to be convinced by my empathy for gays or blacks- I care about trans people themselves! I'm saying if an issue doesn't matter- according to the trans people in this thread- and it's a losing issue, why are we insistent on fighting it anyway? Who benefits from this?
Is there a reason to believe this is how things work? Rather than "within a democracy as long as 'trans rights' means something that most people disagree with, it will be harder to advance the pro- side of the issue"? Is there evidence of 'momentum' being a real thing in politics, as opposed to the 'pendulum' that I've witnessed over the last decade?
But this is not throwing millions of people under the bus. At least according to this thread, there are fewer trans women in sports than bills stopping them. If this issue does represent throwing literally millions of people under the bus- driving them to extinction-, why are you saying upthread that it's niche and irrelevant?
Yeah, I don't get this. If they're niche irrelevant issues that affect ~nobody, why not just concede them? Republicans could hardly dominate the discourse with these topics if there was nobody to discourse against.
If I have something you want and I could give it to you but instead charge you money, is that malice? If so, what system wouldn't make malice indistinguishable from business decisions? If not, what distinguishes that from a login wall in exchange for viewing content?
I don't follow. If I'm a biologist interested in studying life, I can't study rhododendra and ignore the rest. I'd learn almost nothing about locomotion, digestion, sexual reproduction, et cetera. But that doesn't tell me that rhododendra aren't alive. I have no problem believing that LLMs aren't an exact reproduction of a mammalian brain- that studying the one will not give you every piece of information you'd want to know about the other-, but I'm looking for a reason to believe the one is intelligent and the other is not.
Hmm, I guess what I mean is- is there an empirical difference between how a model of intelligence is limited and how a model of language is? Something I can evaluate in a year and say "Oh, the evidence still points to the latter" or else something that could falsify your theory? Otherwise this seems to be a distinction without a difference; believing it should have no effect on my actions or predictions.
On alignment, too, there doesn't seem to be a difference. For a decade I have expected models of intelligence to fall over on alignment. That these purported models of language do the same is hardly evidence that they are not intelligent.
If you only mean there is an undecidable philosophical difference, fair enough. I'm not especially interested in that question.
If a thin model of language can write poetry, perform arithmetic, perform logical reasoning, develop software, play chess, beat factorio, identify and exploit novel security issues, and solve millenium prize problems, what is the purpose of the distinction? Are there tasks that you believe models of intelligence could do that thin models of language cannot?
"Recursive self-improvement"- models making better models
Did you respond to the wrong comment?
What does the one have to do with the other? If I tell you I committed a crime and an independent investigator tells you I committed a crime, why would you need to know my income and outlays to decide whether I can commit crimes?
What does verbal violence mean non-hyperbolically to you?
>Verbal abuse (also known as [...] verbal violence [...]) is a type of psychological abuse that involves the use of oral or written language directed to a victim. Verbal abuse can include the act of harassing, labeling, insulting, scolding, rebuking, or excessive yelling towards an individual.
https://en.wikipedia.org/wiki/Verbal_abuse
>even invents concerns to downplay as if Jasmine had reported on them. In particular, you said that noise is a non-issue – that's not discussed at all in the interview (CMD+F "noise" – zero results).
Huh? How about CMD+F on the comment I replied to? My comment was intended to argue against that, not to summarize a source. I cited the interview towards A.) blue collar workers are getting a good deal B.) localities are getting a lot of tax revenue in return C.) people don't oppose data centers out of fear for their jobs D.) people are primarily upset over secrecy caused by NDAs E.) those NDAs also prevent governments from fighting misinfo.
Separately, I argued that the electricity price, water, and noise issues are not actual problems. I didn't intend to imply that the interview proved that, and, in fairness to you, I could have made that more clear.
Per your objection to C. above, poor phrasing on my part. I meant "not top-of-mind".
>they had real concerns like grid strain, gas generation[...], ugly facilities, temporary jobs, subsidies, secrecy, and an unequal distribution of the benefits.
I didn't say these things were false. I mentioned pollution and subsidies as real, and I highlighted secrecy as the primary issue. Otherwise, I was responding to the parent's claims. My point was never that every complaint one could make about data centers is unfounded; it was that the complaints I was replying to were (except for job loss, which I do believe is a legitimate concern, but which I don't think many agree with me about).
>I find it strange that you're putting more stock in a substack journalist from the Bay area who went to Wisconsin to figure out why people are doing something, than the people who live there and are doing that thing.
??? How is it wrong to trust a person who interviewed people about what those people believe over someone who does not even claim to be involved? How about, instead of a substack, she provide any reason at all we should take her word over someone who talked to people who lived there? As far as I can tell, you take the commenter as the preferred source over the journalist because she agrees with you and lets you say "bingo". You can find my epistemics as strange as you like, they sure beat that.