(Mis)generalization of Helpful-only Fine-tuning
Mirrored from arXiv — Machine Learning for archival readability. Support the source by reading on the original site.
Computer Science > Machine Learning
Title:(Mis)generalization of Helpful-only Fine-tuning
Abstract:Helpful-only models, that is, models that are trained to always follow user intent, are valuable for dangerous capability evaluations and other areas of AI R&D where refusals would be an obstacle. Little is known about the generalization properties of helpful-only training: helpful-only models refuse less than their harmless counterparts, but previous work has not studied other dimensions of their alignment. We study the shortcomings of existing helpful-only models. We find that some show emergent misalignment, others have residual refusal behaviors, and most show poor steerability, sycophancy, and incoherent character. We show that simple anti-refusal training can cause many of these issues. None of these problems are necessary consequences of helpful-only training, though: we show that synthetic document fine-tuning and adding character-related questions to SFT and RL can mitigate them.
| Comments: | 77 pages, 50 figures |
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2606.04413 [cs.LG] |
| (or arXiv:2606.04413v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2606.04413
arXiv-issued DOI via DataCite (pending registration)
|
Submission history
From: Mohammad Omar Khursheed [view email][v1] Wed, 3 Jun 2026 03:43:08 UTC (2,315 KB)
Access Paper:
- View PDF
- TeX Source
References & Citations
Bookmark
Bibliographic and Citation Tools
Code, Data and Media Associated with this Article
Demos
Recommenders and Search Tools
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
More from arXiv — Machine Learning
-
Early Detection of Alzheimer's Disease Using Explainable Machine Learning on Clinical Biomarkers: A Multi-Class Classification Study Using the Alzheimer's Disease Neuroimaging Initiative (ADNI) Dataset
Jun 4
-
Novel Aspects of IEEE SA P3109 Arithmetic Formats for Machine Learning
Jun 4
-
Position: Deployed Reinforcement Learning should be Continual
Jun 4
-
Pseudospectral Bounds for Transient Amplification in Coupled Gradient Descent
Jun 4
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.