A long ramble first, then back to the task: one LLM workflow that keeps working

sull · x · 2026-07-22

A user says a long, casual ramble session works well with LLMs.

Their trick is to pause, go big-picture in a more conversational mode, and then bring the insights back to the task. They say this improves performance, especially when they want the model to better understand the goal before returning to the details.

Related event: Karpathy Shares LLM Tip: Feed Context with 10-Minute Voice Rambling(5 posts)→

Original post →

More from coding & agent

coding & agent channel →