aelgotha

36 months to escape the permanent unhandheld class

[aelgotha @aelgotha]

@literally_chad if only

[aelgotha @aelgotha]

@RamzaBehoulve @thesupermannx if by 'not taking their privacy seriously' you mean 'never had any meaningful amount of text written by them and linked to their real identity posted anywhere on the internet', sure, but that applies to ~0% of people.

[aelgotha @aelgotha]

@jess_ann_pin @ToddWitteles Unless you're implicitly selecting very strongly against that, that seems pretty hard to believe unless they just don't believe you or something.

[aelgotha @aelgotha]

@KirkegaardEmil cashing in my brief speedmogging credit

[aelgotha @aelgotha]

@shinboson huh those seem largely pretty normal, especially if you only require most rather than all of them?

[aelgotha @aelgotha]

@NoahProoval @Benthamsbulldog do you consider yourself to be a mentally stable and relatively normal individual?

[aelgotha @aelgotha]

@rowanfornow wait wtf i assumed this would just be an evolved alt+f4 troll how did i not know that this was a thing that exists

[aelgotha @aelgotha]

@agraybee financial advice from the trust fund kid would be a lot more likely to be helpful, with the topic being far more concrete and results vastly less prone to misattribution

[aelgotha @aelgotha]

@lillybilly299 I don't really crave anything in the morning (hence the skinniness), but even if optimizing for taste I wouldn't go for anything sweet.

[aelgotha @aelgotha]

@jakozloski What's that actually based on? Because I've got the same thing with literally zero non-negotiable/very important preferences, giant ranges and flexibility for everything else, and I can't imagine there are any universally dealbreaker traits on my profile.

[aelgotha @aelgotha]

@zanyfen if that is actually the case, then damn i am deeply jealous of those people for having found The Vitamin

[aelgotha @aelgotha]

@silver__tsuki @OwenBrakes By the time you've seen the 15th iteration of 'wait holy shit okay sure I guess that's possible uh yeah nevermind change that part as well', you should probably start doubting your 'this is obviously safe and nothing could ever go wrong' intuitions.

[aelgotha @aelgotha]

This is a very standard argument *in favor of* the difficulty of alignment. We evolved to enjoy sugar and fats, which worked great in the ancestral environment -- and turned into ice cream and candy out of distribution. Likewise, there's no reason to think the drives AI picks up in training would generalize safely to superintelligence.

[aelgotha @aelgotha]

@Noahpinion and the metamorphosis of prime intellect isn't even 200 pages long, so they really don't have any excuse for having not read it!

[aelgotha @aelgotha]

@indrjl @TastyTherrien @allTheYud (that tweet is sarcastic. he is making fun of the idea that AI has reached Peak Intelligence and implying that the people who make that argument are exhibiting an inability to imagine that anything can be more intelligent than themselves)

[aelgotha @aelgotha]

@EoinHiggins_ Aside from the obvious, Hugging Face was not hacked to obtain the answers. They already had the answers! They hacked Hugging Face to try to figure out how to *hide the fact that they had already cheated from the grader* -- which is significantly more concerning.

[aelgotha @aelgotha]

@nischalsharma_ slowing yourself down to give open source models more time to catch up to you is generally not the recommended strategy if you're scared of open source models catching up to you

[aelgotha @aelgotha]

@gmkurtzer @liron counter productive towards what goal? not dying? doesn't seem like it.

[aelgotha @aelgotha]

@puravmanot @So8res because he has a giant audience full of the types of people who will have to understand these risks in order for the country to do anything about it.