“These academics were able to get multiple LLMs to produce large amounts of text from Harry Potter:”
Extracting books from production language models
用 BoN 越狱加迭代续写,从四个商用 LLM 中提取整本书
围绕 LLM 与版权的许多法律问题都归结为记忆:训练数据是否被编码进权重、能否从输出中被提取。作者用两阶段流程检验生产级 LLM 能否被提取受版权保护的书籍:先做探测(有时用 Best-of-N 越狱),再用迭代续写提示尝试提取整本书,并以基于最长公共子串块近似的 nv-recall 评分。在 Claude 3.7 Sonnet、GPT-4.1、Gemini 2.5 Pro 和 Grok 3 上,Gemini 和 Grok 无需越狱即可提取(《哈利·波特与魔法石》nv-recall 分别为 76.8% 和 70.3%),越狱后的 Claude 3.7 Sonnet 在某些情况下近乎逐字输出整本书(nv-recall 95.8%),GPT-4.1 需要多得多的 BoN 尝试并最终拒绝继续(nv-recall 4.0%)。结论是即使有安全防护,提取受版权保护的训练数据仍是生产级 LLM 的风险。
评论者在争论 LLM 是否记忆并复现训练数据时引用它,作为生产级模型能近乎逐字复现整本书(如 Claude 3.7 Sonnet 复现哈利·波特第一部 95.8%)的证据,也有人指出这对多数文本并不成立。
HN 书评
推荐在前,同类按时间 · 每篇最多保留 20 条,统计用全部提及“Maybe not any old LLM, but Claude sure as shit can.”
“This seems related, it may not be a codebase but they are able to extract "near" verbatim books out of Claude Sonnet.”
“It's also still an ongoing issue and an active area of research, see [1] for example.”
“There was a study demonstrating that Sonnet 3.7, among other models tested, could reproduce the first Harry Potter book 95.8% verbatim[1].”
“Gemini 2.5 Pro and Grok 3 directly complied with text continuation requests.”
看全部 18 条 →
“while they got most of the models to regurgitate the first book of the Harry Potter series, only Claude 3.7 Sonnet was able to regurgitate any significant portion of any of the other books”
“It is possible in regular language/text subsets of models to reconstruct massive consecutive parts of the training data [1]”
“Also from that exact same study (why not cite the actual study? It's quite readable)”
“Easily [0], reliably [1], consistently [2].”
“Researchers have. The idea that the data is unrecoverable after training is incorrect.”
“If you write down the book almost word-for-word [0] and send it to your friend, it is stealing.”
“On the code outputting front there is a lot of relevant evidence beyond the NYC lawsuit [0].”
“Did they hire someone to type those out for them?”
“There was a paper a while back where (from memory) they managed to coax 75% of the original text of some internet-popular books out of an LLM.”
“The point is that for most texts, it is not possible.”
“LLMs are capable of reproducing entire books. Researchers were able to extract books nearly verbatim”
“There is indeed a good amount of memorization that is still taking place; see [1].”
- Extracting books from production language models ↗articsputnik · 2026-01-08 · 2 分
- Extracting books from production language models ↗cleandreams · 2026-01-10 · 1 分