7 pointsby dhorthy6 hours ago1 comment
  • Bolwin3 hours ago
    Two things

    1. If we're using native harnesses, I'd have preferred you use kimi code, not opencode 2. The variation in the two kimi providers just shows how you can't trust n = 1 trials