Posts
Comments
I am glad to see somebody make the point properly. It's a weird state of affairs. We know the models can implement PoCs for CVEs better than most coders. We know the models can persuade people pretty effectively. Obviously the models can spread and change very easily. It's also easy for a rogue deployment to hide because datacenter GPUs draw 70W idle and update scripts constantly use tons of bandwidth. There's just no urgency to any of it.
How do you measure results?
If you wanted to take this idea to an absurd level, you could install a dropped ceiling made partially of furnace filters, and a grid of fans above it. Maybe have the outer perimeter of fans blowing up and the inner area blowing down, to try to get one large convection through the entire room.
How do you figure out the optimal filter thickness? If you hypothetically had a very weak fan then it wouldn't push much air through even furnace filters. If you had a magic constant air flow source then you would want the thickest filter possible.
I guess I am just wondering if you could use something better-looking and cheaper, like semi-transparent paper with lights behind it or a washable sheet/tapestry.
Have you heard of Big Ass Fans? It's a company that makes what you would expect. Do you think your ceiling fan filter could work with a 30ft fan?
What is the current popular (or ideally wise) wisdom wrt publishing demos of scary/spooky AI capabilities? I've heard the argument that moderately scary demos drive capability development into secrecy. Maybe it's just all in the details of who you show what when and what you say. But has someone written a good post about this question?
Einstein started doing research a few years before he actually had his miracle year. If he started at 26, he might have never found anything. He went to physics school at 17 or 18. You can't go to "AI safety school" at that age, but if you have funding then you can start learning on your own. It's harder to learn than (eg) learning to code, but not impossibly hard.
I am not opposed to funding 25 or 30 or 35 or 40 year olds, but I expect that the most successful people got started in their field (or a very similar one) as a teenager. I wouldn't expect funding an 18-year-old to pay off in less than 4 years. Sorry for being unclear on this in original post.
I don't have a witty, insightful, neutral-sounding way to say this. The grantmakers should let the money flow. There are thousands of talented young safety researchers with decent ideas and exceptional minds, but they probably can't prove it to you. They only need one thing and it is money.
They will be 10x less productive in a big nonprofit and they certainly won't find the next big breakthrough there.
(Meanwhile, there are becoming much better ways to make money that don't involve any good deeds at all.)
My friends were a good deal sharper and more motivated at 18 than now at 25. None of them had any chance at getting grants back then, but they have an ok shot now. At 35, their resumes will be much better and their minds much duller. And it will be too late to shape AGI at all.
I can't find a good LW voice for this point but I feel this is incredibly important. Managers will find all the big nonprofits and eat their gooey centers and leave behind empty husks. They will do this quickly, within a couple years of each nonprofit being founded. The founders themselves will not be spared. Look how the writing of Altman or Demis changed over the years.
The funding situation needs to change very much and very quickly. If a man has an idea just give him money and don't ask questions. (No, I don't mean me.)
Wasted opportunity to guarantee this post keeps getting holywar comments for the next hundred years.
This is pretty inspiring to me. Thank you for sharing.
The other day I was trying to think of information leaks that a competent conspiracy couldn't prevent, regarding this. I just thought of one small one: people will sometimes randomly die or have their homes raided. If the slavery is common, then sometimes the slaves will be discovered during these events. Even if the escapees wanted to silence the story out of shame, cops would probably gossip to the press.
So you can probably tally such events, crunch the numbers, and get a decent conspiracy-resistant estimate.
As a layman, I have not seen much unrealistic hype. I think the hype-level is just about right.
You should not bury such a good post in a shortform
Maybe it should be a game that everyone can play
Yeah I just wanted to check that nobody is giving away money before I go do the exact opposite thing I've been doing. I might try to tidy something up and post it first
I do think I could put a good team together and make decent contributions quickly
I can only find capabilities jobs right now. I would be interested in starting a tiny applied research org or something. How hard is it to get funding for that? I don't have a strong relevant public record, but I did quite a lot of work at METR and elsewhere.
I wonder if anybody has tried to quantify how much it's worth to be a swing voter. I imagine if you are the government contractor up for renewal then it's worth quite a lot, but I wonder how much of the money/benefits the average Joe sees.
I don't know much about swing state benefits except that Milwaukee, Wisconsin got their lead pipes replaced by the fed and the workers were required to be local and they say they were paid quite well https://youtube.com/watch?v=4VpwgG0P8VU
Aw man we used the same word for different things again
Your examples fit the definition quite well. Apparently this is in the dictionary now. https://www.merriam-webster.com/dictionary/gaslighting
Regarding this
Such as the moms in the abusive partners example above: each one could acknowledge her self-deception once it was safe for her abusive partner to know too. She got enough power (financial or social) to protect herself and her child, making the telepathic scan no longer a dire threat.
I would add that most abusive people don't really like crushing their loved ones and it is sometimes easy to get them to stop, eg by having a peer of the abuser get a private word with the two parties separately. I think it is common for there to be simple miscommunication/misunderstanding — the abuser does not typically actually benefit from the accusative situation.
Why haven't abuser & abusee already talked and figured this out? Well there is some force field where you can't have a normal conversation with someone who is hitting you (or you are hitting) about the hitting. Although I don't know how to put it in your terms here from this post.
What gaslighting goes on in math class?
I am impressed with how far you thought this through. Amend the constitution, including the constitution amendment section
The opposing states in the coalition will simply declare war against the defectors. It's surely worth keeping your own army to keep being a swing bloc.
I want this to be a board game
I don't know if this would be good for the country, but it would certainly be good political entertainment.
Doesn't everybody always code in a strong time-discount? I have never seen code without it.
The o1 calculation is correct! https://math.stackexchange.com/a/1264753
.5 * (1 - sqrt(1.5e11^2 - 6.4e6^2)/1.5e11) = 4.55e-10
I am surprised. I have seen it mix up million and billion when calculating how many nukes the solar energy that hits earth is equivalent to.
Of course the sun is not nearly a point but whatever.
No good science without some good fun.
Wrong link? Looks like this is it https://arxiv.org/abs/2409.06927
Here is my understanding. Is this right?
Incredible!! I am going to try this myself. I will let you know how it goes.
honesty vector tuning showed a real advantage over honesty token tuning, comparable to honesty vector steering at the best layer and multiplier:
Is this backwards? I'm having a bit of trouble following your terms. Seems like this post is terribly underrated -- maybe others also got confused? Basically, you only need 4 terms, yes?
* base model
* steered model
* activation-tuned model
* token cross-entropy trained model
I think I was reading half the plots backwards or something. Anyway I bet if you reposted with clearer terms/plots then you'd get some good followup work and a lot of general engagement.
Hey!!! Thanks for replying. But did you or anyone you know consider chemical cisgenderization? Or any mention of such in the forums? I would it expect it to be a much stronger effect than eg joining the military. Although I hear it is common for men in the military to take steroids, so maybe there would be some samples there.... I imagine taking cis hormones is not an attractive idea, because if you dislike the result then you're worse off than you started.
(Oh and we were still together then. LK has child now, not sure how that affects the equation.)
Thank you! Seems like this bot works quite well for this task
I have used a number of discourse forums and they just feel bad/wrong but I cannot explain why. I would also vote for more of an old-fashioned php BB with a nice theme. Those are always great, even though all my intuitions tell me they seem like they should suck. Shows how little I know.
Eg https://github.com/phpbb/phpbb
Also has styles: https://www.phpbb.com/customise/db/styles/board_styles-12?sid=6245508b90fd3410be19888406fae215
Basically I'm repeating what Said said
If you have a clear metric to judge candidates on (eg engagement on a linkedin ad) then you might be able to do a super effective and quick performance-based hiring method. Shameless plug: https://www.lesswrong.com/posts/3AZkXwcCJZc5CAFQN/how-to-hire-somebody-better-than-yourself
Good luck!
Thanks for the cached explanation, this is similar to what I thought before a few days ago. But now I'm thinking that an older-but-still-youthful mouse would be better at avoiding predators and could be just as fertile, if mice were long lived. So the food & shelter might be "better spent" on them, in terms of total expected descendants. This would only leave the disease explanation, yes?
Where has the "rights of the living vs rights of the unborn" debate already been had? In the context of longevity. (Presuming that at some point an exponentially increasing population consumes its cubically increasing resources.)
Hey thanks much for sharing new info with me. What a nice comment to read. I was sure someone would come by and be pissed and mean as hell, but folks have been engaging in quite good faith.
but I'm more reserved
I think this might point at the central problem with my evidence. People vary in how publicly they live their lives by orders of magnitude. It could be that only 1% of math geniuses are trans women but they post / get views on Twitter 100x more. Or a similar thing in high school and the workplace. Math professors tend to live quiet lives...
Anyway, unfortunately I think this post might be kinda too toxic/hurtful for the average reader to be worthwhile overall (although nobody has mentioned that to me) and I'll probably move it to a pastebin or something.
I think the basic question (whether hormones are fucking or helping your brain long-term) is quite important and deserves a better treatment. I might try to do that eventually.
Any ideas?
They keywords are much appreciated. That second link is only from 2022! I wonder if anybody suggested this in like 1900. Edit: some of the citations are from very long ago
maybe you die young so you don't get your descendants sick
I've always wondered why evolution didn't select for longer lifespans more strongly. Like, surely a mouse that lives twice as long would have more kids and better knowledge of safe food sources. (And lead their descendants to the same food sources.) I have googled for an explanation a few times but not found one yet.
I thought of a potential explanation the other day. The older you get, the more pathogens you take on. (Especially if you're a mouse.) If you share a den with your grandkids then you might be killing them. Also, if several generations live together, then endemic pathogens stick with the clan much longer. This might eventually wipe out your clan if one of the viruses etc has a bad mutation.
If you die before your offspring even hatch then you might not pass them any pathogens. Especially if you swim a mile up a river that's dry 90% of the year. https://youtube.com/watch?v=63Xs3Hi-2OU This is very funny and 1 minute long.
Most birds leave the nest (yes?) so perhaps that's why there's so many long-lived birds.
Although IIRC, bats live a really long time and have a mountain of pathogens.
Anybody know if this explanation is fleshed out somewhere, or know a better explanation?
Not that it matters, but the new version sounds kind of like the author is making a big deal, and the old version sounds like normal press language, to me at least
Yeah I guess that's another prereq. I think you can make up for it some by having good work for people to do. I would rather work on a cool or valuable thing with amateurs than something lame with pros.
This is great!! I've just been using it now, thanks to your pointer.
I don't know if you're a woman, but the women I know have had much more success in politics than the men I know.
Very valuable information! I have never heard of this artificial mucus stuff. Thank you for sharing.
Why do you want to go the FDA route? You think the gene therapy is the most promising?
JITs are very underrated. Bad JavaScript code runs so much faster than bad C code. Sorry you got downvoted.