Repository navigation
recall: end the reply with what to try when the hits miss - #245
Merged
Merged
Conversation
After a miss, agents rephrase, raise k, follow get lines or read whole sessions, and nothing in the reply says which pays. Measured on the 116 misses of the 244-question recall benchmark (huggingface/funes-memory at 347686a), as answers found and tokens per follow-up: - rephrase in the hits' words, k=8: 51 found, ~3.5k tokens - raise k to 16: 37 found, ~6.6k tokens (the first 8 come back) - follow all 8 get lines: 8 found, ~30k tokens - get the session most hits come from: 40 found, ~180k tokens So the reply now closes with one line: rephrase first, then raise --k, and a get line only reads around its own hit. "no results" keeps its first line and adds the advice to rephrase. ask's prompt is unchanged, since it renders the bare hits for an agent with no tools. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
dacorvo
changed the title
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
After a miss, agents typically:
Nothing in the reply says which pays, but we measured all four strategies on the 116 misses of the 244-question recall benchmark (huggingface/funes-memory in funes-recall-benchmark), as answers found and tokens per follow-up:
So the reply now closes with one line: rephrase first, then raise --k, and a get line only reads around its own hit. "no results" keeps its first line and adds the advice to rephrase. ask's prompt is unchanged, since it renders the bare hits for an agent with no tools.