Metraipolis › archive › dc_12dfe93053f5
Mirror, Mirror on the Wall: Prompt Echoing in Small Instruct Language Models — arXiv:2609.15045
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.15045
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-15T22:53:32.804Z
- SHA-256 of the archived text
a3e4a65b7bbdc8f5e3cc674cda1783c306c29a9d5400fe95cc6f0b205bc72ee8- Stable URL of this page
https://metraipolis.com/documento/dc_12dfe93053f5
How to cite
Mirror, Mirror on the Wall: Prompt Echoing in Small Instruct Language Models — arXiv:2609.15045 — arXiv (en), https://arxiv.org/abs/2609.15045 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_12dfe93053f5 (Metraipolis archive dc_12dfe93053f5, sha256 a3e4a65b7bbdc8f5e3cc674cda1783c306c29a9d5400fe95cc6f0b205bc72ee8)
BibTeX
@misc{dc12dfe93053f5,
title = {Mirror, Mirror on the Wall: Prompt Echoing in Small Instruct Language Models — arXiv:2609.15045},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.15045},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_12dfe93053f5; sha256 a3e4a65b7bbdc8f5e3cc674cda1783c306c29a9d5400fe95cc6f0b205bc72ee8},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.15045. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Prompt echoing is a recognized failure mode of instruct language models, in which a model instead of generating a response, mirrors the provided prompt, even though it did not receive a specific instruction to do so. Is this phenomenon a sign of the model leaking the content of its training dataset, or is it rather caused by a misaligned behavior of the internal induction/copying mechanisms? We investigate prompt echoing small language models from different families (Gemma, Llama, Qwen, SmolLM and OLMo) and show that echoing prompts are likely to have partial overlap with the training dataset but the phenomenon is primarily driven by the model's induction heads. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-15 in cs.AI). Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2609.15045 — 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 2026-10-15T22:29:03.785Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-15?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).