Train the Model, Not the Reader: Decodability Supervision for Verifiable Activation Explanations

  • 类型:arxiv
  • 标识:2607.20379
  • 链接:https://arxiv.org/abs/2607.20379
  • 主分类:multimodal
  • 形态:method
  • 被引:0
  • 被引来源:Semantic Scholar + OpenAlex
  • S2被引:0
  • OpenAlex被引:0
  • 影响力被引:0
  • TLDR:Two audit protocols, the comparison of grounding and truth and the swap to an independent evaluator, and RECAP (Readable Encodings via Co-trained Auxiliary Predictors), linear heads trained alongside the target model to keep designated content decodable are contributed.
  • OpenAlex ID:W7170190251
  • OpenAlex DOI:10.48550/arxiv.2607.20379
  • DOI:10.48550/arxiv.2607.20379
  • DOI来源:OpenAlex
  • 开放获取:green
  • 开放获取链接:https://doi.org/10.48550/arxiv.2607.20379
  • OpenAlex更新:2026-08-24
  • 待LLM分类:否
  • 标题中文:训练模型而非读者:用于可验证激活解释的可解码性监督
  • TLDR中文:提出两项审计协议——grounding and truth 对比以及 swap to an independent evaluator,以及 RECAP(Readable Encodings via Co-trained Auxiliary Predictors),即与目标模型联合训练的线性头,用于保持指定内容的可解码性。
  • 来源文件
  • /inbox/tom/_candidates/2026-07-23-agent-rag-longcontext-candidates.json
  • /inbox/tom/_candidates/2026-07-24-agent-rag-longcontext-candidates.json
  • [S2 enrich]
  • [OpenAlex backfill]