大模型水印系列文章汇总

写在前面

大家好，我是刘聪NLP。

今天早上刷到一篇大模型水印相关论文《Three Bricks to Consolidate Watermarks for Large Language Models》，发给群友们。

结果群友们竟然找到了大模型水印系列文章汇总的Github，特此分享给大家。

          
https://github.com/hzy312/Awesome-LLM-Watermark  

      

大模型水印系列论文

• Advancing Beyond Identification: Multi-bit Watermark for Language Models

          
https://arxiv.org/abs/2308.00221  

      

• Three Bricks to Consolidate Watermarks for Large Language Models

          
https://arxiv.org/abs/2308.00113  

      

• Towards Codable Text Watermarking for Large Language Models

          
https://arxiv.org/abs/2307.15992  

      

• A Private Watermark for Large Language Models

          
https://arxiv.org/abs/2307.16230  

      

• Robust Distortion-free Watermarks for Language Models

          
https://arxiv.org/abs/2307.15593  

      

• Watermarking Conditional Text Generation for AI Detection: Unveiling Challenges and a Semantic-Aware Watermark Remedy

          
https://arxiv.org/abs/2307.13808  

      

• Provable Robust Watermarking for AI-Generated Text

          
https://arxiv.org/abs/2306.17439  

      

• On the Reliability of Watermarks for Large Language Models

          
https://arxiv.org/abs/2306.04634  

      

• Undetectable Watermarks for Language Models

          
https://arxiv.org/abs/2306.09194  

      

• GPTs Don’t Keep Secrets: Searching for Backdoor Watermark Triggers in Autoregressive Language Models

          
https://aclanthology.org/2023.trustnlp-1.21/  

      

• Watermarking Text Data on Large Language Models for Dataset Copyright Protection

          
https://arxiv.org/abs/2305.13257  

      

• Baselines for Identifying Watermarked Large Language Models

          
https://arxiv.org/abs/2305.18456  

      

• Who Wrote this Code? Watermarking for Code Generation

          
https://arxiv.org/abs/2305.15060  

      

• Evading Watermark based Detection of AI-Generated Content

          
https://arxiv.org/abs/2305.03807  

      

• Robust Multi-bit Natural Language Watermarking through Invariant Features

          
https://arxiv.org/abs/2305.01904  

      

• Are You Copying My Model? Protecting the Copyright of Large Language Models for EaaS via Backdoor Watermark

          
https://arxiv.org/abs/2305.10036  

      

• Watermarking Text Generated by Black-Box Language Models

          
https://arxiv.org/abs/2305.08883  

      

• Protecting Language Generation Models via Invisible Watermarking

          
https://arxiv.org/abs/2302.03162  

      

• A Watermark for Large Language Models

          
https://arxiv.org/abs/2301.10226  

      

• Distillation-Resistant Watermarking for Model Protection in NLP

          
https://arxiv.org/abs/2210.03312  

      

总结

大模型水印技术感觉是可信AI的关键，后面估计会有一大批文章。

请多多关注知乎「刘聪NLP」，有问题的朋友也欢迎加我微信「logCong」私聊，交个朋友吧，一起学习，一起进步。我们的口号是“生命不止，学习不停”。PS：交流2群已经成立，欢迎加入。

往期推荐：

• 浅谈LLM的长度外推
• 打造LLM界的Web UI
• 是我们在训练大模型，还是大模型在训练我们？
• Llama2技术细节&开源影响
• 大模型时代-行业落地再思考
• 大模型幻觉问题调研
• 垂直领域大模型的一些思考及开源模型汇总
• 如何评估大模型-LLMs的好坏？
• 阿里「通义千问」大模型-内测分享
• CORGI-PM：首个中文性别偏见探索和缓解数据集
• ACL2022|NoisyTune：微调前加入少量噪音可能会有意想不到的效果
• 总结|Prompt在NER场景的应用
• NAACL2022-Prompt相关论文&对Prompt的看法