The evaluation of the Reasoning Trajectory Detection task will not be con-ducted inside the TIRA sandbox.
⚠️ PAN 的招牌是 Docker 沙箱内可复现评测（1,100+ 次提交都走 TIRA），但唯独这个推理轨迹检测任务不进沙箱，参赛者只提交输出文件，可任意使用外部 API、云 GPU 和开源大模型，评测时不设任何算力与代码约束。结果的可复现性因此明显低于 PAN 其他任务。要把「鉴定推理归属」往制度化（学籍、学分、职称）方向推，第一届的证据强度就先打了折。
we plan to curate additional human-written reasoning trajec-tories with final answers from websites like chegg
⚠️ 藏在细节里的方法学问题：所谓「人类推理轨迹」的金标准，来自 Chegg 这类作业答疑网站。这些解题步骤是被平台格式化、面向应试写出来的，未必代表人类自然推理；而且 Chegg 上的内容近年本身已大量混入 AI 生成。用它当「人类」正类，训练出的检测器学到的可能是「Chegg 文体」而不是「人类思维」。这直接削弱「能鉴定推理归属」的可推广性。
the submitted watermarking systems can be used in a much broader context to authenticate any type of text, not only machine-generated text.
🔴 对提纲「无人能建正向认证」的一个反例，必须先接住：PAN 2026 的水印任务刻意设计成对已有文本加水印，作者明说它可用于认证任意文本、不限于机器生成文本。技术上确实有人在做「给文本盖章」。但要看清它认证的是载体——这段文字来源可追、未被篡改——仍不是「这段思考出自人脑」。正向认证的对象是文本，不是心智。这个区分不讲清楚，台上会被水印一句话反驳。
which is why many AI companies now embed invisible watermarks into the output of their LLMs.
非共识：组织方自己承认生成式 AI 检测有天花板——这话出自连续办了三届 AI 文本检测评测的团队，分量不同于外部批评。更关键的是产业路径已经转向：与其在接收端事后判断「像不像 AI」，不如让模型在生成端就打水印。这意味着未来「是不是 AI 写的」将主要由厂商决定，而不是由学校和期刊检测出来——教育机构在这条链上是没有话语权的一方。
Since PAN 2012, more than 1,100 submissions have been made this way via the TIRA experimentation platform
⚠️ 提纲若想引用检测器准确率，本文查无此数：全文只报告参与规模（2025 年 112 份系统提交、70 篇 notebook 论文；2007 年以来 82 项 shared task；2012 年以来 1,100+ 次提交），一个 baseline、一个历年最佳分数都没有给。这是任务征集公告而非结果报告。要谈准确率必须另找 PAN 历年 overview 的结果章节——用这篇论文说「检测器准确率如何」是引错了文献。
the participant systems should identify the source of the reasoning trajectory and final answer—whether they are generated by an AI system or written by a human.
口径：任务形式是二分类（AI 生成 vs 人写），既不是程度回归，也不是给某个具体的人出具证明。它回答的是「这段推理不像 AI/像 AI」，而不是「这段推理确实是这个学生想出来的」。提纲第 2 题「负向认证 vs 正向认证」的区分在这里拿到了字面支持：连 2026 年最前沿的推理轨迹任务，也只做到把样本归入「人类」这个类别，而非归到某个具名个体。
Finally, we introduce Reasoning Trajectory Detection as another new task in 2026, which has the goal of attributing reasoning trajectories to LLM or human authors and to classify their safety.
✅ 与提纲「PAN 2026 新增推理轨迹检测，把推理轨迹归属为 LLM 或人类作者」几乎逐字对上。关键限定：本文是 2026 年 2 月发布的任务预告（extended abstract），任务尚未开跑，全文一条参赛结果都没有。所以「鉴定一段推理是谁做的已成正式研究领域」可以说，「已经能做到」不能说——目前只有题目，没有答卷。
The 2025 edition was extended with another subtask to determine the degree of human-AI collab-oration in a text, which also received many submissions, but will not return this year.
✅ 提纲称「PAN 2025 设判定文本中人机协作程度子任务」属实。但提纲漏掉了后半句：这个子任务 2026 年不再举办。也就是说，最接近「量化一段文本里人出了多少力」的公开评测，办了一届就停了。提纲用它论证「行业正在建协作度量尺」，实际情况是这把尺子被收起来了——对第 2 题「这套验法能规模化成制度吗」的追问，这反而是更硬的弹药。
SynthID embeds digital watermarks directly into AI-generated images, audio, text or video.
补一条结构性判断：水印天然只能由生成方在生成时刻嵌入。这意味着认证权归模型厂商所有，而人类的思考没有「生成方」可以在源头盖章。想反过来用水印证明「这是人想的」，等价于要求为每一次人类书写配一个可信的记录装置——那是监控，不是认证。第 2 题的两难就在这里。
But it can be hard to tell the difference between content that’s been AI-generated, and content created without AI.
✅ 提纲的第三条论点在这句上得到印证，且比提纲说得更彻底：Google 自己把问题定义成「区分 AI 生成的内容与非 AI 生成的内容」，整套方案的方向只有一个——给机器产出打标。全页没有任何机制去证明某段内容出自人的思考。负向认证是水印能做的极限，正向认证在这条技术路线上根本不是一个目标。
We are currently collaborating with journalists and media professionals to test the portal and collect their feedback.
⚠️ 成熟度限定：验证门户目前仍处于与记者、媒体从业者协作测试、收集反馈的阶段，页面上挂的是「早期测试者候补名单」链接。这不是一个已普遍可用的公共设施。把它描述成「行业已建成 AI 内容认证基础设施」是明显的放大。
Earlier this year, we launched the SynthID Detector, a verification portal, to verify if a piece of content was watermarked with SynthID. Just upload an image, video or audio file.
专业级验证门户同样只收图像、视频、音频文件，文本再次缺席。另注意时间表述是含糊的 Earlier this year，本页整页无日期——这也是为什么提纲那句「2026 年 5 月宣布 5000 万次」无法在此页落地。引用时间线请另找带日期的官方博客。
Simply upload the image, video or audio clip to your chat, and ask if it’s been created or altered by Google AI.
🔴 对第 2 题最有杀伤力的一条：面向普通人的检测通道只接受图像、视频、音频——文本被明确排除在外。也就是说，教育场景里最需要判断的那类内容（一段文字），恰恰是这套官方工具在消费端不受理的。「AI 检测器能管住作业」在官方页面上就已经落空了。
It’s inaudible to the human ear, and can’t be altered by common modifications like adding noise, MP3 compression, or changing the speed of the track.
口径：音频段用了绝对语气 can’t be altered，且限定在 common modifications（加噪、MP3 压缩、变速）。范围之外的攻击——重录、重合成、局部拼接——不在承诺内。另注意音频水印的适用面同样窄：仅 Lyria 生成或 NotebookLM 播客功能产出的音频。
It’s added the moment content is created, and designed to stand up to modifications like cropping, adding filters, changing frame rates, or lossy compression.
注意措辞是 designed to stand up to（设计上能抵抗），不是 has been shown to withstand。本页没有给出任何检出率、误报率、样本量或攻击强度。作为营销页可以理解，但把它当成鲁棒性证据引用是不成立的——「设计意图」和「实测结果」之间隔着整篇论文。
SynthID adjusts these probability scores to generate a watermark. It's not noticeable to the human eye, and doesn’t affect the quality of the output.
🔴 最关键的空白：文本水印的原理是调整下一个 token 的概率分布，而本页对文本水印的鲁棒性一个字都没说——不谈改写、不谈翻译、不谈截断、不谈用另一个模型转述。对比之下图像和音频段落都逐项列了抗干扰能力。这个不对称的沉默本身就是答案：token 概率里的统计印记，经不起一次彻底改写。
We’ve expanded SynthID to watermarking and identifying text generated by the Gemini app and web experience.
⚠️ 文本水印的覆盖范围比一般人以为的窄得多：只限 Gemini 应用与网页版产生的文本。不是所有 Gemini 输出、不含 API 调用、不含 Gemma 等开放权重模型。开放权重模型上的水印在技术上也无法强制——谁都能拿掉那段解码逻辑。这一条直接决定了「靠水印做全网 AI 检测」不成立。
The watermarks are embedded across Google’s generative AI consumer products, and are imperceptible to humans – but can be detected by SynthID's technology.
口径：覆盖范围写的是「Google 的生成式 AI 消费级产品」，不是全部 Google 模型、也不含 API 与开发者链路。所以「Gemini 输出都带水印」这个说法在本页上站不住——本页只承诺消费级产品线。⚠️ 提纲待核项「OpenAI 图像已嵌入 SynthID」在本页零踪迹，本页只谈 Google 自家产品，不能拿它当跨厂商采纳的证据。
SynthID is our new watermarking tool, designed specifically for AI-generated content. It empowers users to identify AI-generated (or altered) content, helping to foster transparency and trust in generative AI.
🔴 三处核验全部落空。本页从头到尾没有出现：Nature 论文（Dathathri et al. 2024）的引用、任何「5000 万次验证」的数字、以及「2026 年 5 月」或任何日期。全文对 Nature、million、Chrome 的检索结果均为 0。提纲这三条都是从第三方转述来的，要用必须另找 DeepMind 官方博客原帖，不能拿本页当出处。
if interrupted mid-task and asked to reflect on its decisions—and never on its actual behavior in the task
✅ 提纲称的 counterfactual reflection training 属实：只训练「被打断时会怎么反思」，完全不训练任务中的实际行为，之后不诚实行为率下降，且 J-space 里亮起 honest / integrity。但口径全空——降幅多少、在哪些评测、跑了多少次，本页一个数字都没有；而且评测是 Anthropic 自己设计、评自己的方法。当「内部思维可被间接塑造」的存在性证据用，别当效应量。
In experiments where we prevented Claude from using its J-space, it still interacted normally, but lost its higher-order cognitive functions.
金句，第 2 题可直接用：说话正常和真在思考是两套机制，前者在后者被切除后依然完好。迁移到「怎么证明你的输出里有人类的思考」时要标清楚这是类比不是证据——本文测的是对模型内部做因果干预，人类写作没有对应的可切除部件。
Without its J-space, Claude speaks fluently, classifies sentiment, answers multiple-choice questions, and pulls facts out of passages roughly as well as before
⚠️「流利保留」的衡量口径原文没交代：只有 roughly as well as before 一句，无困惑度、无人评、无基准分。另外「切除」的操作定义是「删掉每个位置上最活跃的那些内容」（removing its most active contents），不是整块子空间清零——消融强度不同，落差幅度可能不同。用这条论证「流利≠思考」可以，但别当成有效应量的实验数据。
multi-step reasoning drops to near zero, and summarization and rhyming poetry-writing performance fall below the level of a much smaller, intact model
口径核验：提纲称「切除 J-space 后多步推理跌到接近零」——✅ 逐字对得上。但必须补口径：本文没给任何评测名、题量、百分比，near zero 是定性描述，量化结果在 transformer-circuits 论文里而非本页。文中举的多步推理例子是心算 3²−2 和「会织网的动物有几条腿」这类两跳检索，颗粒度很小。上台别说成「在 GSM8K/MMLU 上归零」，那是本页支撑不了的。
The results on more recent models may be confounded by the presence of information about the evaluation in the pre-training corpus.
🔴 提纲完全没有引用的一条脚注，却是全文最重要的自我限定：近期模型在 agentic misalignment 上拿满分，可能是因为这套评测本身已经进了预训练语料——模型见过考题。用提纲第8题的语言说：Anthropic 自己承认，它无法排除自家最新模型是在「刷题」。任何拿「Claude 已满分」论证「教原理有效」的说法，都被这条脚注卡住。
Those whoused their English to find work in hospitality and tourism were waiters or receptionistsor started their own businesses that cater to English-speaking persons. English languagecapability has enabled some female return migrants to bypass traditional domestic serviceand find work as English teachers or move to tourist towns where they can demand a highersalary because of their language skills
I suspect knowing the language can actually go both ways as far as in demand skills for countries
Migrants listed hard-to-measure personal achievementsand competences such as initiative, responsibility, self-confidence, follow-through,punctuality, and presentation of self, along with a number of social skills,
Basically being a good and hard worker translates
Men reportedthe transfer of construction, carpentry, and automotive repair skills; women reported foodand beverage preparation skills and some support and managerial skills such as computerand data entry knowledge.
This ones to answer q2
Thus, some entered entry-level jobs asdishwashers and helpers, while others entered those same industries as cooks and skilledmasons.
Same sample, different outcome?
hey were morelikely than men to discuss not only the technical skills they acquired in their jobs, includingcooking, cleaning, and caregiving, but also social competences, such as team work andintergroup communication skills. Their jobs as receptionists, secretaries, domestics, andcooks made them good candidates for similar positions in the US
Gendered aqcuisition
happiness and suffering
Sobre Se pesquisar aqui e em outras plataformas, achará bastante conteúdo interessante. Se considerar que vale a pena pode me seguir em vários locais (links abaixo), estou produzindo material que considero útil em outros lugares mais adequados, inclusive farei algo que muitos pedem para aprender programar corretamente. Me segue para ficar sabendo quando rolar.
19022023 230626 1-050 R15. SL<br /> o Lido
o The Best!
Coffee’s early connectionwith costly visions of high fashion and Oriental exoticism would persistthrough the eighteenth century, even as its price dropped, familiaritywith it spread through different levels of society
While we can't say anything yet about direct causation,
שאלה חשובה נוספת:...?