Vecr

15 ms·

211.
by Vecr 2y ago
'Cause A caused both B and C? If you see the codex and the shroud contain the "same" information, doesn't it seem reasonably likely they got it from the same place? And people weren't bad at copying back then, so th
212.
by Vecr 2y ago
No it's not, even with only 4 embryos to select you'd generally expect at least one to be fine.
213.
by Vecr 2y ago
In IQ? Assuming reasonable training, nutrition, and lack of brain damage? How?
214.
by Vecr 2y ago
That's disputable, but it would be a really bad idea for obvious reasons either way. Think the incel problem is bad now?
215.
by Vecr 2y ago
Well, yes, but the old "clean up the gene pool" propaganda wasn't ever going to work, and it's not going to work on a practical level now. Gene pool stays almost as bad as it always was, and each generation has to put in
216.
by Vecr 2y ago
Yes, but practically speaking if you constrain to the set of people who would go through with assortive mating in today's society, AND are high IQ (and don't think it's a fluke, i.e. they think it's heritable), AND have
217.
by Vecr 2y ago
Okay then that's pretty much a single-generation effect in practice. A lot of people don't understand that by the early to mid 1930s it was realized that the "propaganda version" of eugenics was not going to work, and if
218.
by Vecr 2y ago
I'm not sure there's really going to be any improvement at all in their descendants, because I really doubt the company is that aggressive. How many embryos did they go through? Much of the genome doesn't run on Punnett squar
219.
by Vecr 2y ago
I'm not sure how well incentivized these organizations are to correctly use the GWAS data vs just appear to do something and then take your money. Assortive mating and then maybe screening for stuff like Tay-Sachs and autism would for
220.
by Vecr 2y ago
I think the theory is doubtful. Sounds too outcome oriented/group-selection-y.
221.
by Vecr 2y ago
It doesn't rely too much on further AI progress. Sure, you'd need a really big training cluster, and an inference cluster, and a supercomputer to put the results together and shuffle the large transformer model I/O, and a goo
222.
by Vecr 2y ago
Already exists, Chasm City by Alastair Reynolds. They absolutely trash orbit by planning very badly and over-packing habitats.
223.
by Vecr 2y ago
Yes, well, it's a cat. They don't have the best mental stability anyway. For the "quick learning" stuff, that's why you would add the more traditional statistical systems. They would be "informed" by the l
224.
by Vecr 2y ago
I've seen similar arguments for a while, and even if it's true now, I'm not sure why it would stay at that level. Even if you assume LLMs fully stop improving, heavy re-training for many customized models (acting as modules)
225.
by Vecr 2y ago
Is it short sighted? If more people want to enjoy leisure and work on hobbies instead of serving others, the ratio will get much worse.
226.
by Vecr 2y ago
That's not the question, the question is the ratio between those-who-want-to-serve and those-who-want-to-be-served.
227.
by Vecr 2y ago
I think KaiserPro is saying authentic human connection doesn't "pay the bills", so to speak. If AI is "about the unreal" as you say, what if it makes everything you care about unreal?
228.
by Vecr 2y ago
I don't put too much stock into complexity classes. They're a real thing for sure, but implementation difficulties and constant factors are too.
229.
by Vecr 2y ago
Breaking crypto, unless that falls too. Not all crypto, I'm skeptical on Grover's. Mostly elliptic curve and integer factorization stuff.
230.
by Vecr 2y ago
I don't see how what you're saying lets you "perceive the whole of the future in one gulp", or maybe it does, but you can't be confidant that it's the real future.
231.
by Vecr 2y ago
You can't exactly predict the future unless you have all the information, even theoretically.
232.
by Vecr 2y ago
You could perceive (maybe? Depends on how it's hooked up) a future (a simulation based the information you have), but there's no reason to think that's what the future is with certainty. Map/territory stuff too.
233.
by Vecr 2y ago
Yes? When I wrote the original comment the actual data looked pretty close, and I still think it's a pretty good match. It would be difficult to re-calculate bridge damage estimates in real time.
234.
by Vecr 2y ago
I got it from the Project Phoenix Scenario.
235.
by Vecr 2y ago
Isn't the standard AI doom position that the government taking control just kills everyone a few years later? The money would be for research, not welfare or whatever the government would use it for.
236.
by Vecr 2y ago
Because you can use money to do things? Even if you don't want private property after AGI, you need it now.
237.
by Vecr 2y ago
It does work, you just need credible trigger thresholds for the salami tactics, treat tactical nukes as strategic, and have enough nukes to punch through ABM.
238.
by Vecr 2y ago
Doesn't matter. He wants to work on AI risk, right?
239.
by Vecr 2y ago
They're generally fine tuned not to. I'm not sure how long that will hold though.
240.
by Vecr 2y ago
Well, it he doing anything to make it go somewhere else? What kind of calculation is he using to figure out who to hit instead?

More ›