– https://t.co/EYOSxBM2Ko
Meta发布了介绍TextSeal2水印方法的论文，论文称该方法在保持生成文本多样性和推测解码接受率的同时提高了水印检测率。
2 reports2 sources
Reports 2
TextSeal2 paper proposes a two-key mixing method for LLM text watermarking
TextSeal2 mixes two keys into one virtual key and uses a different sampling method to accommodate speculative decoding. The cited account reports that, on Qwen3.5-9B at equal text diversity and draft acceptance rates, it detected 89% of 200-token answers at a 0.1% false-positive rate, versus 71% for TextSeal and 67% for textGrain.
Meta paper introduces TextSeal2 to improve AI text watermark detection
Meta's TextSeal2 blends two keys into one virtual key to embed a detectable watermark in AI-generated text. In a test on Qwen3.5-9B using speculative decoding at equal text diversity, it detected 89% of 200-token answers at a 0.1% false-positive rate, compared with 71% for the earlier TextSeal.
SourceRohan Paul (@rohanpaul_ai) ↗Rohan Paul (@rohanpaul_ai) ↗