Xiao Lin

[Xiao Lin @much_science]

@burny_tech You've invented Genesis-O algorithm

[Xiao Lin @much_science]

@rebelcrayon Sutton bench 1 RSI 0

[Xiao Lin @much_science]

@_arohan_ It's more the second order effect, investors anticipation change and then funding dries.

[Xiao Lin @much_science]

@__tinygrad__ Ready for 10T local models

[Xiao Lin @much_science]

@xeophon Struggles to fit in my SSD let alone memory. Take notes local LLM folks!

[Xiao Lin @much_science]

@karpathy @shostekofsky "Intelligence is ten million rules"

[Xiao Lin @much_science]

@RichardSSutton Well superior intelligence doesn't seem to guarantee an understanding of the bitter lesson

[Xiao Lin @much_science]

@WhiskeyPeaky @Silicon_Fly Making AMX more available for local LLM inferences would be higher impact imo. Posing it as a M Ultra/Spark alternative would give better positioning.

[Xiao Lin @much_science]

@realmemes6 Are we past the "new paradigm" stage yet? Was waiting on that one.

[Xiao Lin @much_science]

@teortaxesTex Even xeons today can do 100T~200T bf16 haven't seen people talking about them.

[Xiao Lin @much_science]

@tydsh The transformer + LoRA part makes explicit symmetry assumptions and easier to study. Gradient descent and the symmetry breaking part is more implicit and probably more devils in the details.

[Xiao Lin @much_science]

@1rreverant @acaleb Then model capacity is limiting. Would love to read some theories on this front.

[Xiao Lin @much_science]

@RuiCarrilho5 Solving for-fun math problems on investor funds is questionable also. It's defensible that they are honing their system for the bigger thing but there's the alternative of an open challenge (token grant + award) instead of an internal push.

[Xiao Lin @much_science]

@LarryGold310597 @Yuchenj_UW An RSA guy with AI is 10^20 RSA guys.

[Xiao Lin @much_science]

@who_is_sane @hsu_steve Yes so they stay happy while being lower paid than colleagues. Probably not much longer?

[Xiao Lin @much_science]

@realmemes6 I'm not an expert on the policy side but on the practical side you are looking at hiring a math person from OpenAI to teach with a 50%/50% arrangement instead of a math prof that's 50% teaching 50% government-funded research, say also AI-assisted.

[Xiao Lin @much_science]

@realmemes6 I don't know who are the elites that make 500k+ in education systems. Time wise it'll at least take time to verify the gains from tech. Many working class people today were educated in an age without computers yet new grads have difficulty finding jobs.

[Xiao Lin @much_science]

@realmemes6 Then education will be repriced. Many lecturers accept low pay because of those aspirational problems for their spare time. AI education is not there yet and you probably still want an ever more capable next generation to fill future jobs.

[Xiao Lin @much_science]

@aran_nayebi It's too early to talk about the soul-in-AI question when the soul-in-animal question is yet to be resolved. Build a dog-bark translator first

[Xiao Lin @much_science]

@fuxinli2 Let's turn this around. You could limit each reviewer to pick up to 3 papers to review, and papers with <3 reviews are desk rejected.