Pasting full chat history wastes tokens and adds no value
A tester compared four methods to give an AI assistant memory. Copying the entire chat history cost 20 to 31 times more tokens than looking up notes, yet performed worse on tasks. The study found that structured note-taking was the most efficient and effective approach, saving costs and improving accuracy.
Key points
- Copying full chat history costs 20–31x more tokens than structured notes but performs worse
- Looking up notes before prompting yields better results at lower cost
- No specific AI model or assistant was tested, but findings apply broadly
The experiment did not name a specific AI model or assistant, but the findings highlight a common issue with token-heavy prompts. The author suggests that AI users should prioritize concise, relevant context over brute-force history pasting to optimize performance and reduce costs.
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Agents & Tools
All →- xAI releases Grok Voice 3.0 for voice agents at $0.08 per minute · 2 src
- Figma launches Specs editor and expands agent tools at India event · 1 src
- Claude in Chrome and BiDi extension serve different automation roles in browsers · 1 src
- KarLex AI releases Graphicle to give coding agents local code graphs and memory · 1 src
- Claude Code shifts from instructions to goals for tasks · 3 src
Comments
via GitHub Discussions