He warns reinforcement learning fails to generalize beyond narrow domains.
There will be an extreme irony if these models really are bound by human generated training data. RL doesn't generalize and is only useful in a handful of areas. And we all loose our skills to something that'll forever be a B+ player.
Users in the replies praised the B+ player framing of AI models for highlighting fast iteration and genius-scale output over slow perfection, while some dismissed the core claim about human data limits as nonsense.
@martin_casado Hey, I'll take a C+ player that can give me results in seconds over an A+ player that takes days. Iteration is more powerful than sheer accuracy, because it gives you more time to think and interact with what you're making.
@martin_casado Time to read scalable oversight papers; they’re fun!
There will be an extreme irony if these models really are bound by human generated training data. RL doesn't generalize and is only useful in a handful of areas. And we all loose our skills to something that'll forever be a B+ player.
@martin_casado Having someone at the head of the distribution managing an infinitely scalable team of B+ players can still do quite a bit.
If one pushes the convex hull of "in distribution" out and then have AIs fill in behind you there is still quite a lot of abundance to be had.
@martin_casado B+ to the smartest human in every field is still basically “country of geniuses in a data center”
@martin_casado synthetic data is already past the human ceiling on math
@benrayfield @martin_casado Is there a proof that Turing completeness is enough to generate human-level and human-like AGI?
@martin_casado on the bright side, as every job turns into a "communicate-your-intentions-to-AI" problem - everyone will get really good at communicating with each other (that many people - especially engineers) suck at today :)
@RealAstropulse Oh yeah for sure. Just don't loose your skill!
@martin_casado i always think about things where test and failure is cheap (eg coding) vs expensive (eg enterprise sales). I wonder if that will drive where RL-trained models become great in the near future.
@martin_casado Dude we already know this isn't true
@empathyx100 @RealAstropulse Dude, don't hate on Astropulse. He does some of the best work on retro pixel stuff on the planet.
@martin_casado This would look like many many professions before no? Highly skilled artisans replaced be mediocre but cheap assembly lines.
Eventually the quality of the assembly line surpassed whatever the artisan could do, but it was decades of process iterations.
@martin_casado Did any other technology generate this level of extreme variance about its own utility?
@martin_casado Dario said it does. But from my pov I don’t see it.
@martin_casado What do you think @yacineMTB ?
@martin_casado And worse; possibly never train young people to even be B+ players
@martin_casado Humans are bound by human generated training data.
@neuralamp4ever @benrayfield No
@martin_casado 100% agree. But this was always a trap in academia too. If you get sucked into administration or something else you rapidly lose your ability to push the frontier or effectively utilize your lab to its full potential.
You need to use the bike to train, not to forget how to run.
@cqkten We do?
He warns reinforcement learning fails to generalize beyond narrow domains.
There will be an extreme irony if these models really are bound by human generated training data. RL doesn't generalize and is only useful in a handful of areas. And we all loose our skills to something that'll forever be a B+ player.
