JevMade Sign in
← Back to guides

JevMade field notes / Measured experiment guide

How to test what an AI assistant should remember

This guide explains a testing tool that helps an AI assistant decide which facts to keep. It checks if a detail matches the original conversation, will be useful later, and keeps the original meaning.

Original by Cairn InkEvaluation

Listen to this guide

JevMade’s plain-English explanation

0:00 /

Our summary

This project is a testing tool that helps an AI assistant decide what information to keep from a conversation. Builders use it to check if a proposed memory is accurate and useful before the software actually saves it, preventing the assistant from remembering false details.

The software uses Jev, an AI tool that chooses from options rather than writing an answer. It checks if a detail matches the original text, if it will be useful later, and if it keeps the original meaning. Based on these answers, a strict rule recommends whether to save, skip, or wait.

This tool only recommends what to save; it does not actually store or delete information. It is useful for people building AI software who want to test memory rules. To test your own text, you must provide a private access code to connect to the service.

Key takeaways

  1. Test if a memory matches the original text, keeps the same meaning, and will be useful later.
  2. Save the AI tool's choices and confidence scores next to the clear rule it followed.
  3. Use your own examples to see how the software handles tricky or unclear statements.

All test results were reported by the project creators and have not been independently verified. The planned study on tricky examples is not yet finished, and the test labels were written by AI.

Repository README

Read the original guide Opens the author’s site in a new tab.

Keep this for later

Sign in to bookmark experiments, guides and videos, and keep notes only you can see.

Continue to sign in

We’ll bring you back to this listing.