Metraipolis › archive › dc_fdf1fd1eb8ee
Self-Generated Feedback Destabilizes Test-Time Training: A Causal Decomposition of Long-Horizon Adaptation — arXiv:2610.05076
- Source
- arXiv
- Original
- https://arxiv.org/abs/2610.05076
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-06T05:26:39.346Z
- SHA-256 of the archived text
bd09fd63d25899915132c98a63b82edd242dc889d1b237e0ee7891d795e0dd72- Stable URL of this page
https://metraipolis.com/documento/dc_fdf1fd1eb8ee
How to cite
Self-Generated Feedback Destabilizes Test-Time Training: A Causal Decomposition of Long-Horizon Adaptation — arXiv:2610.05076 — arXiv, https://arxiv.org/abs/2610.05076 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_fdf1fd1eb8ee (Metraipolis archive dc_fdf1fd1eb8ee, sha256 bd09fd63d25899915132c98a63b82edd242dc889d1b237e0ee7891d795e0dd72)
BibTeX
@misc{dcfdf1fd1eb8ee,
title = {Self-Generated Feedback Destabilizes Test-Time Training: A Causal Decomposition of Long-Horizon Adaptation — arXiv:2610.05076},
howpublished = {arXiv},
url = {https://arxiv.org/abs/2610.05076},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_fdf1fd1eb8ee; sha256 bd09fd63d25899915132c98a63b82edd242dc889d1b237e0ee7891d795e0dd72},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv, https://arxiv.org/abs/2610.05076. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Test-time training (TTT) lets a model store information in its weights during inference. When the model learns from its own output, however, each update also changes the model that generates the next training example. Across 128K-token streams, retaining generated-text updates worsens prediction on independent human-written text with three TTT-E2E model configurations (labeled 125M, 760M, and 3B). The same failure occurs when Adam updates Qwen3-4B's existing weights. The same update mechanisms can improve on real text, so writing itself is not the failure. Three matched comparisons trace the causal pathway. Fixed Generation removes over 98% of the damage at 125M and 760M by using a frozen model to generate training chunks. Recorded Replay separates the loss caused by reading degraded text from the additional loss stored by updating on it. A paired one-update comparison then shows the local conflict: an update predicts its source better but new real text worse. This cost grows after Closed Loop adaptation, with a few trajectories accounting for most large failures. Finally, Settlement evaluates the candidate state on independent real text before commitment. It leaves mean endpoint gaps of 0.07 and -0.02 nats at 125M and 760M while retaining real-text adaptation. These results motivate checking prediction on independent evidence before retaining an update. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2610.05076 — 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 2026-10-06T05:24:05.964Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).