Teortaxes responded that organizations should prioritize basic research instead.
@teortaxesTex @FeltSteam they should not have given up until reaching 100T parameters, at least
human brain has 100T synapses and huge sparsity for a reason
@FeltSteam That was the guess but the question is, again, why didn't they bother? it was clearly broken. Token repetitions, pretty meh uplift for the cost They gave up on it
Negative users reacted to the claim that AI should scale to 100 trillion parameters like the human brain by dismissing implementation difficulties as a mere skill issue, which comes across as insulting.
No Digg Deeper questions have been answered for this story yet.
@EgeErdil2 @FeltSteam they should have gone back to basic research maybe they thought they have more time given their RL first mover advantage
@teortaxesTex @FeltSteam they should not have given up until reaching 100T parameters, at least
human brain has 100T synapses and huge sparsity for a reason
@teortaxesTex @FeltSteam it's possible they were betting on massive generalization from RL i suppose
@teortaxesTex @FeltSteam if you can't make it work, the existence proof obviously shows it's a skill issue on your part
@teortaxesTex @FeltSteam they should not have given up until reaching 100T parameters, at least
human brain has 100T synapses and huge sparsity for a reason