研究问题与实验设置
Activation interventions can make a language model produce a target answer, but can the model use the intended condition in other computations? We study this question in Qwen3.5-27B on tool-use tasks where time or permissions determine eligibility and selection.
主要结果
Across 32 fresh worlds, two case-specific residual-writing variants raise familiar target-selection correctness from 35/128 to 119/128 and from 33/128 to 114/128.
边界条件与持续性
- Yet these gains disappear when familiar facts and identifiers are recombined within the same selection task, while selection of previously rewarded identifiers increases.
- The gains persist across output formats but yield little improvement in independent judgments about whether a candidate is eligible or should be selected.
内部编辑对判断的影响
Internal edits can nevertheless change how a source fact affects a judgment: in a prespecified secondary comparison, an early-layer source patch changes 50/64 judgments when that source is used, with one adverse change among 64 cases where it is not.
结论与启示
- This selective effect does not ensure correct complete actions.
- Together, these results separate target-selection control, condition-sensitive judgments, and complete-action correctness.
- Evaluating how an edited state is used, rather than only its trained answer, exposes these differences.