Guardrails removed spam, off-topic, unclear, or duplicate replies.
Ask a question below.
Published answers will appear here.
what if by far the most important way AI "recursively self improves" is just by letting you build RL environments... really fast? would be an explanation why "distillation prevention" totally fails to prevent fast catch-up; why we've had speed up, but without continual learning
@1a3orn I can believe this is a big part of near-term AI R&D automation gains. Expect it is kinda weaksauce relative to how AI systems could self-improve “later in the game”.
nondramatic, therefore not focused upon by companies or AI safety
Guardrails removed spam, off-topic, unclear, or duplicate replies.
Ask a question below.
Published answers will appear here.