“Some an opus and fable subagents, have them each do the estimates separately, have them each argue for their preferred numbers, and then have the main Claude running the analysis make a final decision.”
LessWrong (Curated & Popular) · "Towards The Autograntor" by Scott Alexander
Play from 18:38
18:38
Audio streams from the publisher’s own feed. Verbatim does not host or copy it.
What was described
use sub-agents (Opus, Fable, etc.) to produce independent forecasts, have them debate, then let the main Claude model make the final decision
Said nearby on the same subject
“‘We have really, really robust forecast models that we use to do all of our buying… we’re never going to be sending out flowers that have been sitting in our warehouse for days.’”
“Give the skill Grant Adversarial Estimate to Claude and ask it to install it in itself Then start a new session with Opus 5.5, Medium Effort and Prompt it with”
“First off, what is the decision? Let's define it. When does it have to be made? What is the cost for making it later as opposed to earlier?... what information do I have to have... who are the other stakeholders”