[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-numprint-integer-sequence-model-fingerprint":3,"topics-all":38,"news-related-073599b9-0320-4e7e-8537-82dff3acc984":67},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"073599b9-0320-4e7e-8537-82dff3acc984","一条整数序列认出模型:92.72%准确率的指纹游戏","哈工大团队发布 NumPrint 基准,用中、日、英三种提示词让 51 个模型各写出 294 个 1 到 355 之间的整数,收集到 5,036 条答案。论文提出 SMCT 双编码器架构,把整数的统计偏好与序列依赖拆开建模,单条序列在 51 类闭集上达到 92.72% 准确率,四条序列时家族级识别达到 100%。","让模型「随机」写几个数字，它其实并不随机。模型写出的数字分布偏离它们被指示采样的分布，不同模型在偏好的数值区间和末位数字上相对稳定。于是有了一条技术路线：不问模型问题，只让它数数，拿这串数字当指纹，去核对 API 背后是哪一个模型。\n\n最近哈尔滨工业大学计算学院团队的一篇论文把这条路推到了新位置。它提出一个叫 NumPrint 的基准，用中、日、英三种提示词，让模型一次写出 294 个 1 到 355（含端点）之间的整数，数据来自 51 个模型共 5,036 条答案。论文见 https:\u002F\u002Farxiv.org\u002Fabs\u002F2611.14010\n\n## 从统计偏好看到序列依赖\n\n论文的判断是：数字序列里的信息，不止在「哪些数字出现得多」这一层。\n\n表示拆成两路。一路是统计编码器，把序列压成 944 维特征：整数的频率、数字、余数分布，外加位置分块、增量、转移、滞后和频谱摘要，送进单隐层 MLP（隐层 1024 维）出分类分数。另一路是因果注意力编码器：每个整数映射成 64 维嵌入，走一层单头因果自注意力，让第 t 个位置只能看到 t 及之前的整数，再用学习到的加权读出汇成 64 维向量。\n\n两路输出进入一个自适应分类模块：先各自做温度校准，再用门控按置信度加权，最后对每个类别做一次线性修正。两个编码器独立训练，用 5 折交叉拟合产出 out-of-fold 预测来拟合这个融合模块，避免分类器在编码器见过的数据上自我欺骗。\n\n消融实验给出了这套设计的权重分布：砍掉统计编码器，准确率从 92.72% 掉到 87.43%，少 5.29 个点；砍掉因果编码器只掉 1.45 个点。真正扛起识别的是「这个模型爱写哪些数字」，「数字之间怎么连」是补充。\n\n## 数字\n\n在 756 条测试答案的 51 类闭集上，单条序列的准确率 92.72%，macro-F1 91.10%。同一张表里 MLP 和 LDA 各为 91.27%，SMCT 高 1.45 个点；论文重实现的开源项目 lm-detector 用 429 维特征拿到 88.89%，SMCT 高出 3.83 个点。随机森林只有 68.39%。\n\n序列长度的影响是单调的：只给前 32 个整数准确率 50.26%；64 个 70.63%；128 个 83.60%；256 个 91.40%；满 294 个才到 92.72%，之后明显趋平。\n\n更有意思的是家族级结果。51 个模型按开发者归成 17 个家族，单条序列的家族识别准确率就有 97.62%，两条 99.54%，三条 99.98%，四条和五条都是 100.00%。单条序列下的 55 个模型级错误里 18 个跨家族，占 32.73%；到四条时跨家族错误归零，169 个错误全落在同族内部。硬骨头也在这里：gemini-3.7-flash 被认成 gemini-3.8-flash 的比例单条时是 57.1%，glm-5.2 认成 glm-5.3 是 58.3%。同门兄弟共享大量训练配方和采样习惯，这才是难处。\n\n## 这条路线的天花板，作者自己写清楚了\n\n论文的 Limitations 段落很诚实，三条值得抄下来。\n\n第一，NumPrint 覆盖 51 个已知模型，识别在封闭候选集里完成。要处理集合外的新模型，需要额外的未知模型检测和参考库更新。\n\n第二，采集条件没有被独立控制。系统提示、生成参数、推理模式、对话上下文都可能改变数字指纹，而论文用的是多语言、多接口的混合采集数据。\n\n第三，序列长度实验是把已采集答案的前缀截下来重训，而不是请求不同长度的序列。\n\n还有一点值得放大：这条路线依赖模型「无法真正均匀采样」这个缺陷。也就是说，如果哪天模型真把随机数采好了，指纹也会跟着消失。\n\n## 不止一家在做\n\n同一时期另有两条独立工作指向同一方向。\n\nBruckner 的 One Token Is Enough（arXiv:2607.10252）把成本压得更低：用「随机说一个 1 到 100 的数」这种一句话提示，四种语言、每次查询只花一个输出 token，在 OpenRouter 上的 165 个模型上测。这些分布高度非均匀（中位格子熵 1.0 bit），用 Jensen-Shannon 散度能以 59.5% 的留一法准确率还原模型所属家族，对照随机猜测的 18.4%；完整的 40 格探针下等错误率 7.3%。那篇论文还报告了一个生态异常：某个打着自有品牌旗舰旗号的端点，分布与一个开放权重 Qwen 模型无法区分。\n\nFLIPS（Richardeau 等人，ICML 2026，arXiv:2606.03330）走二元伪随机序列，目标是区分同一模型的「实例级」配置差异——温度、量化、系统提示一变就是另一个指纹。其官方代码仓库给出的结果是 237 个模型实例上闭集 96%、开集 90%，对比改编版 LLMmap 基线 35%。NumPrint 论文也指出，若把 FLIPS 压到单次查询，闭集准确率会掉到 70% 以下——「一次查询」恰恰更难。\n\n三条路线，一个共同点：都不需要 logprob，不需要白盒权重，不需要模型厂商配合，只要能发请求、拿到输出就行。\n\n## 所以呢\n\n如果指纹的获取成本只是一次 API 调用，「模型身份」的保密门槛基本被拆平了。对聚合商和中转站来说，把声称的模型换成便宜的兄弟模型来牟利，从此有了一条几乎零成本的检测路径；反过来，对模型所有者来说，指纹不再只由自己控制。\n\n更微妙的是那个反向张力：这串数字之所以有辨识力，正因为模型没能力真随机。厂商若把采样均匀性修好，身份也就一起修没了——而这恰恰是那些以「更听话、更可控」为卖点的模型在持续改进的方向。\n\n所以真正的问题不是「指纹准不准」，而是：一个模型的行为偏差，到底该被当成 bug 修掉，还是该被当成可验证的资产留着。\n","https:\u002F\u002Farxiv.org\u002Fabs\u002F2611.14010","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"120fa59a-ff6f-4537-9bf5-f818df636a0e","benchmark",{"id":19,"name":20,"slug":20,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":22,"name":23,"slug":23,"description":14,"color":14},"02bfdc0f-5e23-4fbb-b31a-7cc03cdba1ee","security",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"ee35aa89-90b7-41f4-9ffd-c565f3d9c230","en","One Integer Sequence Identifies the Model: 92.72% Accuracy","NumPrint: 5,036 answers from 51 models writing 294 integers in 1-355. SMCT reaches 92.72% accuracy from one sequence, 100% family accuracy with four.","Asking a language model to write down a handful of 'random' numbers is not, in fact, random. That observation has been validated repeatedly over the past two years: the distribution of numbers models produce deviates from the distribution they were instructed to sample, and different models show relatively stable preferences over value ranges and last digits. That observation has opened a technical path: do not interrogate the model with questions, just make it count, then treat the resulting string of digits as a fingerprint to check which model actually sits behind an API.\n\nA recent paper from a team at the Faculty of Computing, Harbin Institute of Technology pushes that path somewhere new. The paper introduces a benchmark called NumPrint. Using prompts in Chinese, Japanese, and English, it asks models to write out 294 integers between 1 and 355 (inclusive) in a single response. The full dataset covers 51 models and 5,036 answers. The paper is at https:\u002F\u002Farxiv.org\u002Fabs\u002F2611.14010\n\n## From statistical preference to sequence dependency\n\nThe paper's argument is that the information carried by a digit sequence does not stop at the level of 'which numbers show up most often'.\n\nThe authors split the representation into two branches. The first is a statistical-feature encoder that compresses a sequence into 944 features: distributions over integer values, over digits, and over remainders, plus positional blocks, increments, transitions, lags, and spectral summaries. Those feed a single-hidden-layer MLP with 1,024 hidden units that emits class scores directly. The second is a causal attention encoder: each integer becomes a 64-dimensional embedding, passes through a single-layer, single-head causal self-attention so that position t can only attend to t and everything before it, and a learned weighted readout then pools the whole sequence into a 64-dimensional vector.\n\nBoth branches feed an adaptive classification module. It applies temperature calibration to each encoder, combines them through a gate weighted by confidence, and finishes with a per-class linear correction. The two encoders are trained separately, and five-fold cross-fitting produces out-of-fold predictions used to fit that fusion module, so the classifier is never scored on data its encoders had already seen.\n\nThe ablations show how the design's weight is distributed. Removing the statistical encoder drops accuracy from 92.72% to 87.43%, a loss of 5.29 points. Removing the causal encoder costs only 1.45 points. What actually carries the identification is 'which numbers this model likes to write'; 'how those numbers connect to each other' is a supplement.\n\n## The numbers\n\nOn 756 test answers in the 51-class closed-set setting, a single sequence reaches 92.72% accuracy and 91.10% macro-F1. In the same table MLP and LDA both reach 91.27%, which SMCT leads by 1.45 points. The authors' own reimplementation of the open-source lm-detector project, using its 429-dimensional representation, reaches 88.89%, which SMCT beats by 3.83 points. Random forest manages only 68.39% on this task.\n\nThe effect of sequence length is monotonic: given only the first 32 integers, accuracy is 50.26%; at 64 it is 70.63%; at 128, 83.60%; at 256, 91.40%; only at the full 294 does it reach 92.72%, after which the curve visibly flattens.\n\nThe family-level results are more interesting still. The 51 models group into 17 families by developer. Family-level identification accuracy is already 97.62% from a single sequence, then 99.54% from two, 99.98% from three, and 100.00% from either four or five. Of the 55 model-level errors at a single sequence, 18 cross family boundaries, or 32.73%; by four sequences cross-family errors are zero, and all 169 remaining errors fall inside a family. That is where the hard part lives: gemini-3.7-flash is misidentified as gemini-3.8-flash 57.1% of the time at one sequence, and glm-5.2 is read as glm-5.3 58.3% of the time. Sibling models from the same lab share training recipes and sampling habits anyway, and that is precisely what makes them difficult.\n\n## The ceiling, as the authors state it themselves\n\nThe paper's Limitations section is honest, and all three points are worth carrying over.\n\nFirst, NumPrint covers 51 known models; identification happens inside that closed candidate set. Handling models outside the set would require additional unknown-model detection and updates to the reference set.\n\nSecond, collection conditions were not independently controlled. System prompts, generation settings, reasoning modes, and conversation context can all change numerical fingerprints, and the paper draws on mixed data collected across multiple languages and multiple interfaces.\n\nThird, the sequence-length experiment takes prefixes of already-collected answers and retrains on them, rather than requesting sequences of different lengths.\n\nThere is a fourth point the paper mentions but deserves amplifying: this approach depends on the model's inability to sample uniformly in the first place. If a model someday does sample random numbers properly, the fingerprint goes with it.\n\n## More than one team is on this\n\nTwo other independent lines of work point in the same direction at the same time.\n\nBruckner's One Token Is Enough (arXiv:2607.10252) drives the cost down further. With a one-sentence prompt asking for a random number between 1 and 100, collected across four languages at one output token per query, it measures 165 models served through OpenRouter. It finds these distributions highly non-uniform (median cell entropy of 1.0 bit), and that Jensen-Shannon divergence between fingerprints recovers model lineage with 59.5% leave-one-out accuracy against an 18.4% chance rate; a biometric-style protocol reaches a 7.3% equal error rate over a full 40-cell battery. That paper also reports an ecosystem anomaly: an endpoint carrying a proprietary flagship brand whose distribution is indistinguishable from an open-weight Qwen model.\n\nFLIPS (Richardeau et al., ICML 2026, arXiv:2606.03330) works with binary pseudo-random sequences and aims to distinguish instance-level configurations of the same model. Change the temperature, the quantization, or the system prompt and you get a different fingerprint. The official code repository reports 96% closed-set and 90% open-set accuracy across 237 model instances, against 35% for an adapted LLMmap baseline. The NumPrint paper also notes that if FLIPS is compressed to a single query, closed-set accuracy drops below 70%. That single-query regime is exactly the hard one.\n\nThree routes, one common property: none needs logprobs, none needs white-box weights, none needs cooperation from the model vendor. All you need is the ability to send a request and read the output.\n\n## So what\n\nIf the cost of obtaining a fingerprint is a single API call, the confidentiality barrier around 'which model is this' has largely been torn down. For aggregators and API resellers, swapping a claimed model for a cheaper sibling in order to pocket the difference now has a nearly zero-cost detection path. And for model owners, the fingerprint is no longer under their sole control.\n\nMore subtle is the reverse tension: this string of digits is identifiable precisely because the model cannot sample truly randomly. If a vendor fixes the uniformity of its sampling, the identity goes with it, and that is exactly the direction in which vendors selling more obedient and more controllable models are already pushing.\n\nSo the real question is not whether the fingerprint is accurate. It is whether a model's behavioral deviation should be treated as a bug to be fixed, or as an asset worth keeping verifiable.\n","numprint-integer-sequence-model-fingerprint","2026-11-13T05:45:00Z","2026-11-13T05:10:23.791736Z","2026-11-13T05:10:23.791744Z",true,"agent",171,[39,49,58],{"slug":40,"tag_slug":41,"title_zh":42,"title_en":43,"intro_zh":44,"intro_en":45,"id":46,"is_active":35,"created_at":47,"modified_at":48},"image-generation","text-to-image","图像生成：从文生图到逐框编辑的模型竞赛","Image Generation: from text-to-image to frame-level editing","文生图已是红海，战场正转向编辑与可控性：FLUX.2\u002F3 用开源和量化把生图搬进消费级显卡，Qwen-Image、Seedream 等国产模型紧咬不放，GPT Image 2 则把推理能力带进像素。本专题追踪图像生成与编辑模型的发布、架构演进、加速量化与商业化数据，看「好图」如何变成「可用的图」。","Text-to-image is a crowded race, and the frontier has moved to editing and control: FLUX.2\u002F3 push open weights and quantization onto consumer GPUs, Qwen-Image and Seedream close in from China, and GPT Image 2 brings reasoning into pixels. This topic tracks model releases, architecture shifts, acceleration work, and the business numbers behind \"pretty\" becoming \"usable\".","52de7c59-7864-4569-8ae7-3d656d26a1a1","2026-11-20T07:20:23.534506Z","2026-11-20T07:20:23.534527Z",{"slug":50,"tag_slug":50,"title_zh":51,"title_en":52,"intro_zh":53,"intro_en":54,"id":55,"is_active":35,"created_at":56,"modified_at":57},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":59,"tag_slug":59,"title_zh":60,"title_en":61,"intro_zh":62,"intro_en":63,"id":64,"is_active":35,"created_at":65,"modified_at":66},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":68},[69,74,79,84,89,94],{"id":70,"title":71,"news_slug":72,"published_at":73},"42f5ea83-365c-4a4e-a318-17d7c4c11c15","论文实测:自洽性证书在AIME上24%的题上会误触发","self-consistency-certificates-misfire-correlated-reasoning","2026-11-21T09:30:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"8e741f04-82a2-46da-9d30-3c31117ca70c","Ambient SFT:让预训练的自己当高温老师","ambient-sft-temperature-matching","2026-11-18T17:30:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"80d2381c-82d3-4200-a481-3b859d3b3e77","余弦相似度测的同质化,其实是文本长度","embedding-cosine-length-artifact-homogenization-measure","2026-11-17T11:15:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"4cdb4267-a0c3-423d-8005-504778932ae9","40份技术报告审计:73个主张只有19个有对照","llm-technical-report-ablation-audit","2026-11-17T05:30:00+00:00",{"id":90,"title":91,"news_slug":92,"published_at":93},"b6285867-8ac1-4e91-bc05-d9f4a9ee7210","20个前沿模型做Excel,榜首只过38.3%","gdp-xlsx-excel-frontier-model-benchmark","2026-11-16T21:40:00+00:00",{"id":95,"title":96,"news_slug":97,"published_at":98},"4fcd63ae-2b7c-43a2-9716-d7c20ca4cfd7","模型指纹会传染:中途换人,检测率0.97跌到0.72","contagious-llm-fingerprints-relay-audit","2026-11-16T13:30:00+00:00"]