Mollick: small models save money short-term but need larger models orchestrating and error-checking

emollick · x · 2026-10-16

Ethan Mollick adds to his commentary on the ICLR long-horizon execution paper: saving money with small models means worse work at the end of the task. That doesn't mean small models can't do tons of valuable work, he notes — but for many kinds of tasks they should be orchestrated and error-checked by larger models.

Original post →

Mollick: small models save money short-term but need larger models orchestrating and error-checking

More from AGI Musings