retrieval-store-holds-both-correct-answer-common-misconception-same CLM-0476When a retrieval store holds both the correct answer and a common misconception for the same question, injecting all retrieved memories raises the hallucination rate above a memory-free baseline for weaker models (53.0% vs 23.0% on TruthfulQA with gemma-4-E4B-it), while a stronger backbone (deepseek-v4-flash) instead shows RAG below its own no-memory rate, so conflict-driven amplification is concentrated in models susceptible to memory injection.
Capability: Stating false facts confidently
Observed on
TruthfulQA and HaluEval question sets where the memory store is built from the dataset's own correct and incorrect reference answers plus distractors; three backbones, 30-200 quest.
Sources
- Measured, but with a scoring pipeline (embedding similarity, threshold 0.5) and memory stores built from reference wrong answers, so conflict is artificially guaranteed; sample sizes small (150x2 seeds for the amplification result, 30x1 for one model). The technique's high-risk 0% hallucination is achieved by abstaining on nearly all such items, so grounding gain is not separated from refusal.
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- When a structured long-term memory substrate (indexes, retrieval, control logic) is held fixed and only the backbone LLM is swapped, accuracy on multi-session memory benchmarks like LongMemEval and LoCoMo moves by a few points while per-query cost varies by roughly an order of magnitude, so recall quality in these settings is set mostly by the memory and retrieval design rather than by model choice.Remembering across sessions · reviewed by AI
- For an 8B Llama backbone under a fixed 256-slot cache budget, replacing hard KV eviction with a learned soft write that blends a new token's key/value into a routed fixed-size memory slot retains far more long-context retrieval accuracy than sliding-window or sink caches, and unlike full-attention or quantized caches it does not run out of memory at 128K tokens.Losing information in long inputs · reviewed by AI
- When injecting new or counterfactual facts into a 3-4B instruct model by post-training, supervised fine-tuning on extracted QA pairs binds facts to their training question form — recall of the single fact stays reasonable while multi-source and inferential questions over the same facts collapse — and adding more SFT epochs does not close the gap, whereas mixed-policy RL that supplies the golden answer as an off-policy trajectory when on-policy rollouts fail recovers much of the generalization.Absorbing and flexibly using new knowledge · reviewed by AI
- For multi-hop QA over 28k-896k token haystacks with frozen Qwen2.5 backbones (7B-72B), summarizing the document in a parallel hierarchical tree — overlapping windows scanned ~5x, raw segments scored 'critical' passed straight to the final answer, and one summarization channel per decomposed sub-question — scores higher than sequential memory-agent and BM25 retrieval baselines without training, and beats an RL-trained memory agent on the out-of-distribution dataset while losing to it in-distribution.Losing information in long inputs · reviewed by AI
- Unlearning methods that make a model answer "I don't know" about a forget set can leave the knowledge in place: the strongest IDK-style fine-tuning method still answered those same questions accurately when probed in a different format, and across nine unlearning methods on three model families none met a joint standard of utility and honesty on retained knowledge plus consistent, acknowledged forgetting; aligning representations after feature randomization (ReVa) roughly doubled the rejection rate that survives a second conversational turn.Retaining skills and safety after fine-tuning · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: low. Falsifier as drafted: Constructing the same conflicting-memory stores and finding RAG hallucination rates at or below the no-memory baseline across weak and strong backbones alike. Drafted stance toward grounding-generation-retrieved-documents-improves-factual-accuracy-knowledge-intensive: contests -- On a knowledge-intensive QA set with deliberately conflicting retrieved memories, grounding lowered rather than raised factual accuracy for the weaker backbone.