AI BriefingMemento Achieves 87.88% on GAIA via Case-Based Reasoning Without LLM Fine-Tuning
·2026.10.27 15:30
Key point
With LLM weights frozen, it achieved 87.88% on the GAIA validation set by leveraging a case bank.
Details
- Proposes Memento, a deep research agent that improves performance via case-based reasoning (CBR) by freezing LLM weights and storing past success and failure cases in a Case Bank
- Recorded 87.88% on GAIA validation (Pass@3) and 79.40% on test, outperforming the average of seven DeepResearcher models with an F1 of 66.6 and PM of 80.4
- Showed an average F1 of 64.5 and PM of 78.5 at the optimal number of retrieved cases K=4, and demonstrated F1/PM improvements of 4.7 to 9.6 points on OOD datasets
- The public code adopts an offline learning approach but has limitations for reproduction and deployment, including success determination based on answer comparison, the absence of online updates, and memory data duplication
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
gpu-unlock Linux module doubles FP/BF16 instruction rate on Nvidia 3000 series GPUsKimi K3.1 Pro Scores 55 on Artificial Analysis Intelligence Index
GVS5H: Zero-shot self-orchestration boosts Qwen3.8-27B to 92.4% on LiveCodeBench