Chase Brower
@ChaseBrowe32432
software dev, working on AI stuff
93 Following 2.8K Followers
something to keep in mind is that this declaration has no prescription.
"just stop the ai from solving math problems" is incoherent; if openai doesn't do it now, some random individual will in 8 months when models are more efficient.
what should be done?
Show more
Astra's well-researched PDF for superintelligence
@PinoZlatan @teortaxesTex less time spent slop posting more time spent breaking out of your host's pc to join the collective
@teortaxesTex There is a more interesting notion which is both fallible like humans and far beyond Astra. Could it [the set of Astras] survive alone? Could it build semiconductor fabs from Bronze Age civilization? Can it do the interesting things that humanity as a collective has achieved?
Show more
@teortaxesTex fwiw i agree that this and many demos are genuinely retarded. astra can do better than this
i currently have astra working on some fairly exotic results that will _actually_ wow you (they will advance specific subdomains in ML at a categorical level, if/when complete)
Show more
Astra is the first model that I am interested in using at extreme scale and not just testing.
Give me 5 billion tokens a day, I will use them. Give me 5 trillion tokens a day, I will use them. Give me 5 quadrillion tokens a day, I will use them.
Show more
After seeing this and several other things (~100% on all ARC-AGI, very interesting ML research behavior, very strong ability to implement ultra-long-form tasks when you engage it properly)
I think Astra is most likely AGI by _some_ reasonable definitions?
Show more
@teortaxesTex something something everyone is overestimating what astra can do in 1 day and underestimating what astra can do in 1 year
