Can you verify a blind watermark without leaking the params? TIM says yes, using Nova+Spartan for the ZK plumbing.
trending8
01 02 PathMark watermarks MoE LLMs by hiding ownership in expert-routing paths, since dense-model tricks keep breaking.03 HeRo watermarks LLM output with hierarchical vocab routing, so verifiers only see the metadata they're cleared for.04 24 bits of watermark in generated code, spread across names and semantics, with proofs and Reed-Solomon keeping score.05 Rethinking Forgery Attacks on Semantic Watermarks in Black-Box Settings: A Geometric Distortion Perspective arxiv.orgPaper models black-box watermark forgeries as geometric distortion on latent manifolds, then uses that to catch them pre-verification.06 LineageMark: Multi-user White-box Watermarking for Contribution Tracing in Model Derivation Chains arxiv.orgWhite-box watermarking for model lineage, with signals designed to survive re-finetuning, re-watermarking, quantization, and pruning.07 ePrint paper on pseudorandom codes from dense-planted LPN, plus first standard-model public-key CCA1 PRC, because parity checks were tired.08 First binary public-key PRC hits 1/6-ε substitutions; the catch is distorted McEliece and subexp security on raw Reed-Solomon codes.