Sam Altmam talked at MIT a few days ago and our president’s first question for him was “What is your P(doom)?”
Imagine this happening two years ago
Asher
527 posts
- I want to get back into college math without it feeling too far removed from the real world. Especially interested in AI-relevant subjects. Any recs?
- What is the ideal time breakdown for a college undergrad? Time spent on classes, research, fun, exercise, etc? Is this even the right question to be asking?
- Called it
I’m gonna be making a lot of money on these AI IMO markets. Math competitions are too narrow to last much longer, and the incentive to work on these systems is too strong - Replying to @arithmoquineAt some point someone told me that intel makes all their chips the exact same way. They're only divided into categories based on how many deformations they have in practice
- I’m gonna be making a lot of money on these AI IMO markets. Math competitions are too narrow to last much longer, and the incentive to work on these systems is too strong
- MIT President Sally Kornbluth opening Sam Altman's talk by asking for his P(Doom) -- full clip "Whether you say its 2 or 10 or 20 or 90, the point is it's not 0" -- @sama
00:00Sam Altmam talked at MIT a few days ago and our president’s first question for him was “What is your P(doom)?” Imagine this happening two years ago - no patrick mckenzie hate but he would not make a good coffee shop ownerThis sounds like a trivial observation and it isn’t: No organization which makes its people pay for coffee wants to win.
- Replying to @devanshpandey and @mattyglesiasthe devansh constant (include 90% error bars on any market devansh can meaningfully influence)
- Every so often I remember u can find the hidden_dim H of any model by feeding it N>H queries and looking at the singular values of the corresponding (N x H) matrix. And also find the umembed matrix up to a rotation by looking at U. And we just figured this out a few months ago.
