一篇新的研究论文挑战了“思考型token”在推理模型中必然会提高安全性的假设。研究发现，像GPT-OSS、Qwen、Olmo和Phi这样的模型的拒绝或合规结果，从第一个token开始就高度可预测，甚至在可见的深思熟虑发生之前。研究表明，“思考”过程更像是前缀补全，结果很少在初始阶段后发生变化，并且目前的干预措施可能通过过度拒绝来无意中压制真正的深思熟虑。 AI
影响 挑战了AI深思熟虑能提高安全性的假设，暗示需要新的方法来诱导真正的安全考量。
排序理由 该集群包含一篇发表在arXiv上的研究论文，讨论了关于AI模型安全性的发现。
在 Hugging Face Daily Papers 阅读 →
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
arXiv:2606.25013v1 Announce Type: cross Abstract: Today's reasoning models use thinking tokens to attain stronger performance on benchmarks than their instruction-tuned counterparts. It is also generally believed that this more "deliberative" mode should improve alignment and saf…
Today's reasoning models use thinking tokens to attain stronger performance on benchmarks than their instruction-tuned counterparts. It is also generally believed that this more "deliberative" mode should improve alignment and safety, by providing the model a safe space to consid…
Research reveals that reasoning models' safety outcomes are predictable from early hidden representations, with deliberation appearing but not substantially influencing final responses, and current safety interventions inadvertently suppress genuine deliberation signals.