vals🔸

philosophy and psychology, understanding and community.
good reasoning and systemized winning enjoyer.

dm me your life problems.

less serious @TutorValsLife

[vals🔸 @ValsTutor]

new vals arc: vulnerability, earnestness, rawness, unfliterdness, ugliness
i'm so used to wanting a good appearance, to choosing to say the things which are good for me.
I'm weak, I have problems, I'm struggling with things.
Give me feedback as u like. I'll continue existing

[vals🔸 @ValsTutor]

@CineraVerinia @MatthewJBar What do you take an intelligence explosion to mean, what are your criteria?

I very much agree that without one, the year superintelligence arrives it would be quite expensive to run, but the idea of the explosion is that we do get recursive improvements in efficiency very fast.

[vals🔸 @ValsTutor]

not my usual but just venting that @NaceAI seems like an unethical company happy to trawl for community slack workspaces then spam thousands of people indiscriminately. mb they should consider not doing that instead.

[vals🔸 @ValsTutor]

@CineraVerinia @MatthewJBar I'm assuming no foom and no nanotech, no incredible technologies. But cheap fast Von Neumann level genius at every human task, 10 billion of them. At marketing, organization, management, logistics. They are calling you and negotiating how to improve your life asap.

[vals🔸 @ValsTutor]

@CineraVerinia @MatthewJBar A year where GDP is doubled is roughly a year where everyone buys twice as many cars, travels twice as much on holiday, receives twice as much healthcare.
Imagine a superintelligence calling u on ur phone offering you all that, u'd be a shmuch not to take its productivity advice

[vals🔸 @ValsTutor]

@robinhanson https://t.co/v9YGwMpvDK

[Tom Chivers @TomChivers]

Sure, this guy is predicting AI doom in the next year, but he was predicting AI doom in ten years nine years ago, and it hasn’t happened yet

[vals🔸 @ValsTutor]

@inerati @SurrealistShip why?

[vals🔸 @ValsTutor]

Fable 5.2 replies: I don’t think anthropic shadow survives the math. Toy case: 50/50 between “90% doom” and “10% doom”, and we make it through. SIA gives a 9:1 update toward the safe world. SSA with all observers as the reference class gives 9:1 too. You only get “no update” if your reference class is survivors only, which assumes the conclusion. (Teruji Thomas, “Dispelling the Anthropic Shadow”; Jessica Taylor, “SSA rejects anthropic shadow, too”.)

The real asymmetry is that doom forecasts only ever get graded in the branch where they were wrong. That’s an unfair game, but a scoreable one. And “intelligence explosion in 2027” resolves well before anyone dies.

[vals🔸 @ValsTutor]

@AtlasOfCharts @NunoSempere Yeah

[vals🔸 @ValsTutor]

With no hate on messier in particular, I didn't wake up a loser, this loser premise makes no sense to me.

(I switched from field building to object level technical ai safety work cuz crunch time etc etc)

[messier ⋆˚꩜.ᐟ 🔎 ⏹ ⟢ @rssmrm]

the intelligence explosion will be here by 2027. get offline, start making art. see your friends. i quit my job and am making music all day. if anyone needs help, ill help. if anyone needs love, i will love. but no longer am i a slave to this system

[vals🔸 @ValsTutor]

@_NathanCalvin It's true we could decide to not handover, but it's unclear that we should, in the worlds where we've successfully solved alignment

[vals🔸 @ValsTutor]

@AtlasOfCharts @NunoSempere Fwiw all LLMs have this problem, including different families from different companies, and the probability that none of the frontier or near frontier companies put $50mil into this seems vanishing. It's a strong prior that it's already been tried and failed that I have

[vals🔸 @ValsTutor]

@GuiveAssadi @KelseyTuoc I generally think it's poor criticism, because I expect most technical ai safety work to accelerate AI overall, since it would have needed to be done at some time anyway, and is being done in advance by smart people on philanthropic money

[vals🔸 @ValsTutor]

@scottleibrand @eli_lifland @peterwildeford +1

[vals🔸 @ValsTutor]

@IsaacKing314 Isaac said ~"There really needs to be ponctuation for a paraphrase"

[vals🔸 @ValsTutor]

@evereverchange @NunoSempere That to me more implies that their math capabilities training includes clearer writing than that it generalized

[vals🔸 @ValsTutor]

@AtlasOfCharts @NunoSempere Fwiw my gripes are more that it's bad at thinking and theory of mind, not specifically style, so a human writer can use heavy assistance to produce a good text if they're guiding and dictating taste. It's autonomous writing I'm sure they're terrible at (again, on my values)

[vals🔸 @ValsTutor]

@AtlasOfCharts @NunoSempere Oh could u link it and some other review you liked? I'm blind here so curious to read both and comment (tho I hope they're not too long)

[vals🔸 @ValsTutor]

@reconfigurthing @herbiebradley @NunoSempere Theory of mind related, like it'll constantly use words it knows but the reader doesn't, like how a project is called internally which obviously the client/other person doesn't know. It like assumes the reader is a copy of itself, very typical mind