Vibeleaderboard
← All Intel
Intel / video

State-Of-The-Art Prompting For AI Agents

Source
youtube.com
Author
Y Combinator
Date

Context

One concrete technique in Y Combinator's Light Cone episode on prompting is to give the model a way out. The hosts relay a finding from a YC startup they work with: a model told to return output in a fixed format tends to give an answer even when it lacks the information, telling you what it thinks you want to hear. The prompt should therefore say what to do when information is missing, such as stop and ask rather than guess.

YC's own variant adds a field to the response where the model can complain that its instructions are confusing or underspecified. Run in production on real user data, those complaints, which the hosts call debug info, become a to-do list for the developer. This is experience reported by the hosts and their startups, not a controlled comparison.

Terms in this piece · Glossary
  • AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
  • context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
Read the source www.youtube.com
More from Y Combinator
Recommended reads
Comments

Checking sign-in…

Loading comments…