Skip to content

Comment on My failed attempt at AGI on the Tokio Runtime

Comments

The only hope I have is to try something completely novel

I don’t think this is true. Neural networks were not completely novel when they started to work. Someone just used a novel piece — the gpu. Whatever the next thing is, it will probably be a remix of preexisting components.

Right. Ironically I chose a model that was around in the 1970s without knowing it.

My point was more a game-theoretic one. There is just no chance I would beat the frontier labs if I tried the same things with less compute and less people. (Of course there is almost 0 chance I would beat them at all.)

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.