Karpathy's 70-minute breakdown of how the top 1% actually work with LLMs
blaizedsouza · x · 2026-11-20
Another retweet variant of the same event: Karpathy's 70-minute course on how the best AI users actually work with LLMs, covering Prompts → Context → Tools → Workflows and how the "top 1%" operate. The companion article "Loops and Graphs" explains how to stop babysitting agents and only approve the final step.
Largely overlapping with the other retweet in this chain; see that summary for detail.
More from coding & agent
- Karpathy: prompting is going away — graphs are the only layer that survives — blaizedsouza · 2026-11-20
- Obsidian Starter Kit v6 bundles your vault into one file for Claude Code with token limits — dSebastien · 2026-11-20
- Fixed-rule scan of 3,631 public MCP servers flags 24 as do-not-use — Minimum-Gift-5217 · 2026-11-20
- Browser MCP benchmark: 176k-565k tokens per bug fix across 5 tools — VisionInBinary · 2026-11-20
- Developer uses Enhanced Radar MCP to monitor his own flight's ATC comms live — massimosgrelli · 2026-11-20
- tinyeval: a Bash-only minimal eval framework with a Claude Code skill — criticasterdotcom · 2026-11-20