Reproducing and Evaluating the Generalizability of Subliminal Learning in Open-Weight Models

2026-09-11

AI Summary

本文通过再现和扩展实验，探讨了潜意识学习在开放权重模型中的普遍性问题，使用新偏好类别、任务及模型验证原论文结论的普适性。

Abstract

In this reproduction paper we investigate subliminal learning, a consequence of distillation where teacher models transmit behavioral preference traits through semantically unrelated data. The original paper explores two types of traits (animal preferences and misalignment), three data modalities (number sequences, code, and chain of thought), and several model families. We reproduce their experiments and extend the setup along three axes: new preference categories (actors and politicians), a new task (chess move generation), and an additional open-weight model (Ministral8B). We also run a controlled ablation on the numbers task's answer-space size (1-, 2-, and 3-digit sequences). We focus on open-weight models with accessible checkpoints on HuggingFace, since the original paper's GPT-4.x fine-tuning is no longer available. Our reproduction supports the original paper's claims, but our extensions show they are not universal as transmission strength varies across traits and tasks, and one model shows almost no effect at all.

Problem

• subliminal learning
• open-weight models
• generalizability
• behavioral preference traits

Innovation

• subliminal learning
• open-weight models
• behavioral preference traits
• controlled ablation
• generalizability