Unverified50% confidenceBenchmarkExact time
The FEB-30 benchmark proposed in the arXiv paper "Beyond Downstream QA" uses an inversion paradigm to construct deterministic ground truth, and tests show that a representative memory system correctly classifies memory operations (create, update, strengthen, supersede) only 20% of the time.
1
Sources
50%
Confidence
Long-term
Relevance
10/19/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn memory tests, retaining the full context allowed the model to accurately locate specific details mentioned by students 95% of the time, whereas after summary compression the accuracy dropped sharply to 32%.69% similarUnverifiedIf 20% of the output fails schema validation, what should be fixed is the prompt or the model tier, not one's mindset.67% similarUnverifiedAccording to the ForgetEval research, when leading memory systems were scored on test cases written by people other than the authors themselves, internal test scores of 63%-68% plummeted to 28%-33%.66% similarUnverifiedIn a synthetic logistic regression classifier experiment, after 80 steps of gradient-ascent-based repair, the unlearning loss rose from 0.0545 to 0.1649, yet the trigger attack success rate remained 100%.65% similarUnverifiedAfter introducing an explicit resolver to handle information supersession relationships, stale-memory failures dropped to zero and overall accuracy jumped from 33% to 78%.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1067939API
curl https://kongchang.com/api/v1/knowledge/claims/1067939MCP
get_claim(id=1067939)