Metraipolis › archive › dc_04ec63443c45
Surprise Alone Learns Lies: Evidence-Gated Continual Learning for LLMs from Streaming Claims — arXiv:2610.39840
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.39840
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-31T06:17:23.678Z
- SHA-256 of the archived text
33fbe6dc13a09dc9e5c3e2cacbcf121492262f66cf5b1e29d896ed36bd120b72- Stable URL of this page
https://metraipolis.com/documento/dc_04ec63443c45
How to cite
Surprise Alone Learns Lies: Evidence-Gated Continual Learning for LLMs from Streaming Claims — arXiv:2610.39840 — arXiv (en), https://arxiv.org/abs/2610.39840 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_04ec63443c45 (Metraipolis archive dc_04ec63443c45, sha256 33fbe6dc13a09dc9e5c3e2cacbcf121492262f66cf5b1e29d896ed36bd120b72)
BibTeX
@misc{dc04ec63443c45,
title = {Surprise Alone Learns Lies: Evidence-Gated Continual Learning for LLMs from Streaming Claims — arXiv:2610.39840},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.39840},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_04ec63443c45; sha256 33fbe6dc13a09dc9e5c3e2cacbcf121492262f66cf5b1e29d896ed36bd120b72},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.39840. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Continual learning methods for large language models increasingly decide what to learn by surprise, keeping the inputs the model finds least predictable. This rule is unsafe when the stream contains lies. A planted falsehood about something the model knows has low probability under the model's belief, so it is highly surprising by construction, and a surprise-driven learner absorbs nearly every such lie, at a rate similar to random selection with the same budget. We study the problem with LoRA-adapted students (Qwen2.5-1.5B, Qwen2.5-3B and Llama-3.2-3B) that learn from a stream of true and false claims about invented people and real CounterFact facts. We report four findings. (1) A verifier key, a judge that must approve each claim, removes lies to the extent that its verifier is accurate; real LLM judges are cautious and reject unsourced truths and even source-backed updates. (2) Calibrated regression tests on unrelated knowledge slow learning down without filtering out lies. (3) The original model's own prior flags claims that contradict what it knows, without a judge. It separates lies about known facts from true new facts (AUC 0.78-0.83 across three models) but not from legitimate updates of the same kind of fact (AUC 0.53). It saves 43% of judge calls, but an attacker with access to the model can choose lies that pass it. (4) If trusted sources exist that an attacker cannot plant, routing contradicted claims to evidence recovers legitimate updates and the combined system has the lowest poison rate among learned claims; when an attacker plants matching sources for 30% of the lies, the system ends up worse than the judge alone. The acceptance signals answer different questions, and none of them decides on its own whether a claim is true. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno). Categorie arXiv: Machine Learning (cs.LG) ; Computation and Language (cs.CL); Cryptography and Security (cs.CR). Pagina del lavoro: https://arxiv.org/abs/2610.39840 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-31T06:07:23.414Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).