Understanding AutoGPT's thought processes

[Photon1c]

From what I understand the "thoughts" that the agent undergoes during the planning phase are "thoughts, reasoning, plan, criticism, speak, and action" i.e see here. However AutoGPT has the following order: "criticism, plan, reasoning, speak, and text".

Am I missing the lines that allow for adjusting these? I can't seem to locate them in agent.py or in the forge folder. I don't like the current order that starts out with criticism, but perhaps there was some "reasoning" behind this :). Thanks!

[ntindle]

This wasn’t written by us but was well written. The current methodology starts with criticisms so that it primes plan for considering the criticism. This could be a flawed design however and I encourage you to swap the order and benchmark it :)