Data scientist at MIRI
Robi Rahman
Karma: 162
In case you haven’t seen,
https://epoch.ai/blog/could-decentralized-training-solve-ais-power-problem
This turned out to be totally wrong. I wish it had been true.
This is wrong in various ways. For example, it’s not strictly worse for liberals if Bob’s RNG makes him play a fascist policy. If he played a liberal policy, liberals have no information about whether 0, 1, or 2 of Alice and Bob are fascists, whereas if he played a fascist policy, liberals know 1 or 2 of Alice and Bob are fascists.
No, (at least for men) it takes much longer than 10 months to make a kid in a way that’s worth doing. You have to find a partner willing to do it with you, which takes an unpredictable amount of time. (I guess if you’re rich you can hire an egg donor and a surrogate.)
Related: youtuber becomes the world’s richest person by making a fictional company with 10B shares and selling one share for 50 GBP
I think the “Successful Treaty” and “Terrorists” scenarios are impossible as written.
There’s too much economic incentive to create AGI. With algorithmic and hardware progress, eventually it will become possible to make an AGI with slightly more computing hardware than a gaming laptop, and then it’ll be impossible to stop everyone from doing it.
Correction: AlphaStar used 6*10^22 FLOP, not 2*10^23. You have mixed up TPU chips and TPU boards.
What does McDonald’s sell in France?
Ingenuous was my first thought too, but it doesn’t have the meaning I think you intended. Perhaps disingenuous?
But if your theory is found to be against the Second Law of Thermodynamics I can give you no hope; there is nothing for it to collapse in deepest humiliation. ― Arthur Eddington
I think you meant “there is nothing for it but to collapse in deepest humiliation”.
It is completely ingenious to mention only Frank for that claim
Ingenious? You mean misleading?
One year later, in light of the ChatGPT internet and shell plugins, do you still think this is just a theoretical thing? Should we worry about it yet?
The fire alarm sentiment seems to have been entirely warranted, even if the plan proposed in the post wouldn’t have been helpful.
Why is the application deadline this month if the retreat is happening next summer?
I’ll be bringing Wits & Wagers, Just One, Secret Hitler, and Innovation.
Lumifer was using the word “expected” correctly.
It’s awful that you were downvoted in this thread when you were mostly right and the others were mostly wrong. I’m updating my estimate of LW’s average intelligence downward.
No it doesn’t! A coin biased towards heads can have p(H) = 0.6, p(T) = 0.4, and each flip can be an independent trial. The total results from many flips will then be Poisson distributed.
They’re Saturdays at 19:00, but not necessarily the 2nd weekend of the month. We kind of just poll the attendees at the end of each meetup and find out when people are available next month.
The meetups are ongoing! We’ve had meetups for the past 50+ consecutive months with no interruptions, although we had to be outdoors and sparsely distanced during quarantine.
Careful not to learn the wrong lesson from this post. Voting really is useless in 99% of races. This one is different: it’s an open seat because the incumbent retired, and there was no one with even double-digit support until relatively late in the cycle. Also there’s a huge difference in the candidates in their effects on x-risk.
Most likely, local elections in Korea don’t matter and it’s probably not worth your time to vote unless you have nothing productive to do.