Metraipolis › archive › dc_10b6fb3efa87
THESIS-MoE: Trainable Hierarchical Extraction and SteerIng of Sycophancy in Mixture-of-Experts — arXiv:2608.15687
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2608.15687
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-18T17:26:02.852Z
- SHA-256 of the archived text
d816a7db29a1423f8bf086a42b12f6d227741cd0943721431df9bbe6c3f52924- Stable URL of this page
https://metraipolis.com/documento/dc_10b6fb3efa87
How to cite
THESIS-MoE: Trainable Hierarchical Extraction and SteerIng of Sycophancy in Mixture-of-Experts — arXiv:2608.15687 — arXiv (en), https://arxiv.org/abs/2608.15687 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_10b6fb3efa87 (Metraipolis archive dc_10b6fb3efa87, sha256 d816a7db29a1423f8bf086a42b12f6d227741cd0943721431df9bbe6c3f52924)
BibTeX
@misc{dc10b6fb3efa87,
title = {THESIS-MoE: Trainable Hierarchical Extraction and SteerIng of Sycophancy in Mixture-of-Experts — arXiv:2608.15687},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2608.15687},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_10b6fb3efa87; sha256 d816a7db29a1423f8bf086a42b12f6d227741cd0943721431df9bbe6c3f52924},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.15687. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Sycophancy, the tendency of a language model to change its answer to match a user's stated belief, is a common alignment failure. Existing activation steering methods typically apply a single contrastive direction uniformly throughout the model, which is an unconditional intervention that alters activations even when no sycophantic behavior is present, trading knowledge retention for behavioral correction. In Mixture-of-Experts (MoE) models, prior work further suggests that behavior is encoded within expert computations rather than routing decisions alone, making precise behavioral steering particularly challenging. In this work, we introduce a shared contrastive signal, built from matched prompts with and without a stated belief, that identifies where sycophancy lives across the MoE hierarchy and drives interventions that act only where the behavior is present. We formulate localization as a causal search over a granularity ladder of MoE blocks, experts, attention blocks, and heads, and compare unconditional subtraction against two conditional alternatives: an analytic projection-based subtraction and a learned per-token gate that steers the model away from sycophancy while keeping its weights frozen. We evaluate on three MoE models measuring sycophancy alongside general knowledge and reasoning benchmarks. Our conditional interventions removed up to 90\% of the belief-induced sycophancy. Our results demonstrate that sycophancy resides in identifiable computational subcircuits and can be selectively steered while maintaining a favorable removal-retention trade-off. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-18 in cs.AI). Categorie arXiv: Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2608.15687 — 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 2026-10-18T16:29:13.820Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-08-18?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).