3 ms·

Isn't that just RL with extra power-intensive steps? (An entire model chugging away in the goal function)

• hrn_frs 9mo ago
  That's correct, but if successful you'd essentially have updated the LLM's knowledge and capabilities "on the fly".
  • ethbr1 9mo ago
    Maybe we could run off-peak load of that nature, when power is cheaper. Call it dreaming. ;)