SLAM: Structural Linguistic Activation Marking for Language Models
📰 ArXiv cs.AI
arXiv:2605.05443v1 Announce Type: cross Abstract: LLM watermarks must be detectable without compromising text quality, yet most existing schemes bias the next-token distribution and pay for detection with measurable quality loss. We present SLAM (Structural Linguistic Activation Marking), a novel white-box watermarking scheme that sidesteps this cost by writing the mark into structural geometry rather than token frequencies: sparse autoencoders identify residual-stream directions encoding lingui
Full Article
Title: SLAM: Structural Linguistic Activation Marking for Language Models
Abstract:
arXiv:2605.05443v1 Announce Type: cross Abstract: LLM watermarks must be detectable without compromising text quality, yet most existing schemes bias the next-token distribution and pay for detection with measurable quality loss. We present SLAM (Structural Linguistic Activation Marking), a novel white-box watermarking scheme that sidesteps this cost by writing the mark into structural geometry rather than token frequencies: sparse autoencoders identify residual-stream directions encoding lingui
Abstract:
arXiv:2605.05443v1 Announce Type: cross Abstract: LLM watermarks must be detectable without compromising text quality, yet most existing schemes bias the next-token distribution and pay for detection with measurable quality loss. We present SLAM (Structural Linguistic Activation Marking), a novel white-box watermarking scheme that sidesteps this cost by writing the mark into structural geometry rather than token frequencies: sparse autoencoders identify residual-stream directions encoding lingui
DeepCamp AI