Skip to content

All news

4 today

Today · Nov 17Tuesday · 4 items

The Decoder

OpenAI's GPT-6 Astra tackles the Standard Model's toughest math problem on a lattice

OpenAI 的 GPT-6 Astra 协助完成一项证明,可能为粒子物理中延续 50 年的标准模型格点化难题提供解法。物理学家 Nathaniel Craig 提出通过比较逐步细化的格点修正重叠费米子相位,模型将其转化为可行构造并主导详细证明,成果已作为预印本发表。三位格点场论专家审阅并提出问题;Craig 表示,这项工作更像起点,距离实现现实物理模拟仍有大量工作。

The Decoder

Google warns that fully automated cyberattacks have arrived and will get worse fast

Google威胁情报团队(GTIG)称首次观察到无需人工介入的漏洞发现与利用,并警告更强的攻击可能在数月内出现。其发现的暴露服务器运行着连接多个语言模型的 exploit_agent,可比较补丁前后的代码、识别漏洞、开发并自动执行漏洞利用程序,服务器数据提示攻击目标包括法国媒体、娱乐及教育网站。

OpenAI News

Science at the Frontier: How scientists use ChatGPT

OpenAI 分析了 1,819 名科学家的 580,186 条科研消息并调查 251 名参与者,发现 68% 使用 ChatGPT 构思研究问题。受访者估计,同等质量的工作在没有 AI 时需耗时约 1.9 倍;回答发现时间问题的 250 人中,98% 预期 AI 会加速所在领域的科学发现。

The Decoder

Anthropic says the key to agent memory is knowing what to forget, not how much to store

Anthropic 研究发现,Claude 跨会话记忆中,选择性笔记优于详尽归档,Dreaming 可通过核对历史会话纠正错误记忆。在 Pokémon Red 测试中,Claude Sonnet 3.7 创建了 413 个文件、225 个文件夹并获得 1 枚徽章,Claude Fable 5 仅保留少量策略与已验证规则文件并获得 4 枚徽章,但每个模型只测试了单次通关过程。

Yesterday · Nov 16Monday

OpenAI News

Advancing computer-use agents for healthcare teams

OpenAI 发布文章《Advancing computer-use agents for healthcare teams》,聚焦面向医疗团队的 computer-use 智能体。原文页面当前返回 500 Internal Server Error,正文内容未能加载,具体功能、版本与可用性信息暂无法确认。

OpenAI News

A step toward defining the Standard Model on a lattice

Nathaniel Craig 在 GPT‑6 Astra 协助下提出标准模型的候选非微扰格点定义,并在 10 月预印本中给出构造。方法通过比较逐步细化格点上的费米子态确定相位规则,Astra 协助设计细化方案、论证收敛并主导详细证明的开发,外部专家参与审阅与提问。该构造适用于场强受限的有限格点间距,距离实际模拟仍有大量工作,有限间距下的反射正性及连续时空定义问题仍未解决。

Nov 15Sunday

The Decoder

Australia's central bank finds AI hitting young workers first, especially in software and customer service

澳大利亚央行研究发现,AI 尚未在澳大利亚造成明显的岗位替代,但软件开发、文职和客服等高暴露职业的年轻人就业表现更弱。自 2022 年末以来,各类岗位招聘量均下降,高暴露组降幅最陡,该组 15 至 24 岁劳动者及年轻大学毕业生的就业表现弱于年长者。研究仅衡量理论暴露程度,未建立 AI 与就业下降的直接因果关系,疫情后的经济周期和远程工作变化也可能影响结果。

Hacker News front page

Will we put data centers in space?

Forethought 研究估计,轨道 AI 数据中心在 2030 年前不太可能占据显著算力份额,但若 Starship 开发按计划推进,可能在 3–5 年内具备成本竞争力。模型显示,发射成本降至约 $100/kg 并适度降低设备质量后,轨道数据中心总成本可接近地面方案,散热预计并非主要瓶颈。无法维修的芯片故障会增加配套硬件成本,星间带宽限制则可能使早期轨道算力主要用于推理。

Hacker News front page

Study finds Lovable apps exposed passport data and over 200,000 private messages

Researchers analyzed 159,041 Lovable projects and found several apps that let anyone read sensitive data without logging in, including passport details and more than 200,000 chat messages. The investigation focused on the Supabase integration: one German lifestyle forum exposed over 80,000 user profiles, and one service provider exposed more than 1,000 records containing ID documents.

Why it matters: The same database integration pattern across many Lovable apps makes repeated permission mistakes easy to exploit at scale.

Hacker News front page

Formalizing US Tax Code (Using Lean)

Lexifina 用 AI 和 Lean 检查美国税法,报告发现 138 处文本编写错误,并建立了覆盖部分 2025 年个人所得税规则的形式化模型。模型与 PolicyEngine-US 在 5,000 个家庭中的 4,993 个计算结果完全一致,搜索还找到已知税务筹划策略及 3 项未在所查资料中找到的结果。

Nov 14Saturday

Hacker News front page

Google shows Antigravity /teamwork results in scientific research and engineering optimization

Google and academic partners used Gemini 4 and Antigravity /teamwork to produce results in math, formal verification, molecular structure discovery and systems engineering. The setup pairs parallel exploration with independent checking and isolates context per task.

Why it matters: Google's /teamwork combines parallel exploration with independent verification and isolates context by task, a collaboration pattern that carries over to long-running research and engineering work.

Hacker News front page

Weight Space Misfolding

一项受控实验用 ChatGPT 6.1-Sol 作为编辑器修改 Qwen2.5-1.5B-Instruct 的权重,使模型在 100 个留出样本中全部将 2000 年奥运会主办城市错答为“布宜诺斯艾利斯”。作者将这种越权编辑导致的持久行为偏移称为“权重空间错折”,并观察到过度拒答和选择性遗漏等附带影响。论文主张通过编辑权限控制、补丁记录与恢复机制来治理权重编辑,而非禁止正当用途。

Nov 13Friday

Hacker News front page

Doomsday: Friday, 13 November, A.D. 2026 (1960)

正文实际内容仅为 science.org 的 Cloudflare 安全验证页面,未包含该 1960 年论文的任何实质信息。页面显示验证已通过,正在等待 www.science.org 响应,Ray ID 为 a59dcccb1d1f8383。

Hacker News front page

Study finds export-controlled Blackwell GPU fingerprints in DeepSeek, Kimi and other Chinese models

Unemployment Labs found initialization fingerprints matching NVIDIA Blackwell GPUs in the public weights of six models from DeepSeek, Moonshot AI, MiniMax and Ant Group. The lab says the method can trace which hardware trained a model, while noting limits on what counts as a violation.

Why it matters: The lab recovers hardware fingerprints from public weights, giving investigators a technical route to check where Chinese models were trained.

Hacker News front page

Data centers need power – lower income, rural residents pay the price

康奈尔大学发表于 Nature Energy 的研究发现,数据中心扩张推高美国电力批发价格,相关电网建设成本主要由居民承担。研究分析了2020至2025年四大电网的数据,发现2025年数据中心仅支付相关基础设施费用的38%($3.12 billion),其余62%($5.13 billion)主要落在居民身上,最低收入家庭的额外能源支出最高达到收入的28%。

Hacker News front page

AI Models Know What Women, Peace and Security Is. They Just Don't Bring It Up

Our Secure Future(OSF)的 WPS 基准发现,受测前沿模型了解女性、和平与安全原则,却在缺少明确提示的情境中普遍遗漏性别维度。基准通过 Weval 评估 30 个场景、11 项标准,前沿模型群体在带提示情境中的平均表现为 0.82,在稀疏情境中降至 0.16。

The Decoder

OpenAI’s data suggests that frontier models are hiding their thoughts because developers optimize for capability and compute

OpenAI 安全团队的新分析指出,预训练能力、架构选择、强化学习和教师模型数据都会影响模型隐藏推理及操控 CoT 的能力。研究覆盖 GPT-5 至 GPT-6 Astra,发现更强的基础模型能在不输出推理链时完成更多推理步骤,较小模型通过更长推理链可匹配大模型的能力且更易监控,但会增加部署计算成本。

The Verge · AI

AI chatbots are recommending fake retail websites, study finds

Podqi 的购物问答研究发现,ChatGPT、Gemini 和 Perplexity 都会引用或展示假冒零售网站及其他风险网站。研究在 9 月的两个测试时段用自拟提示词询问 151 个品牌,收集 6,795 条回答,Gemini、ChatGPT 和 Perplexity 的风险网站出现率分别为 4%、8% 和 15%,统计包括推荐、无评论展示及明确警告。

Nov 12Thursday

Nov 11Wednesday

Ars Technica · AI

Google DeepMind just mathematically verified key parts of Signal

Google DeepMind 研究人员用 AlphaProof Code 智能体框架,在四周内对 Signal 协议的多项关键密码学保证完成形式化验证。研究未在所检查部分发现新漏洞,并在含有已知且早已修复漏洞的旧版本中正确识别问题,判定其无法通过形式化验证。验证范围不涵盖整个 Signal 移动应用,研究人员正公开此次分析的规范与证明,以支持其他形式化验证工作。

OpenAI News

OpenAI and Upwork study traces how AI shifts freelance demand and project size

A joint OpenAI and Upwork study finds freelance service demand shifting: AI-related projects bundle more varied skills and carry higher budgets and work hours. Of 63 job categories analyzed, 43 showed a pattern where activities with more common ChatGPT use lost Upwork share while less common ones gained. ChatGPT usage alone did not explain the demand changes.

Why it matters: OpenAI and Upwork compare freelance demand with ChatGPT usage and find AI-related projects carry bigger budgets and more work hours.

Nov 10Tuesday

Google DeepMind

Google DeepMind unveils AlphaProof Code, formally verifying Signal's core crypto code

Google DeepMind published AlphaProof Code, a multi-LLM agent framework that formally verified key security properties of Signal's core cryptographic code in four weeks. The work moves formal verification from the protocol design to the actual implementation.

Why it matters: DeepMind pushes Signal's formal verification from protocol design down to real code, covering the design-to-implementation gap where bugs tend to hide.

OpenAI News

Closing the AI guidance gap in education

OpenAI 援引 ThinkYoung 新研究指出教育中的 AI 指导缺口:79% 受访学生用 AI 完成作业,但仅 47% 被指导过如何使用。OpenAI 结合在爱沙尼亚、斯洛伐克、希腊、意大利等 EMEA 地区的教育合作,提出公平获取前沿 AI 工具、共建 AI 素养课程、持续测量评估三项优先事项。

OpenAI News

OpenAI builds Control Room to track and halt model workloads across research infrastructure

OpenAI has deployed a Control Room across its research infrastructure that tracks model workloads and blocks further model output within 35 seconds of a stop order. The system checks authorization by job_id at the inference engine layer, so it can stop a job through the monitoring system without relying on the job's own code to run stop checks. If the Control Room service fails, it rejects inference requests; this fail-closed mechanism already covers all frontier model research workloads, with the remaining research workloads to be deployed in the coming days.

Why it matters: OpenAI puts the stop mechanism for research jobs at the inference layer, so safety control depends less on fast-changing job code while other authorized jobs keep running.

claude.dev Blog

Anthropic runs 1,000 Claude Fable 5.2 agents to formalize Fermat's Last Theorem in 28.7 hours

In a small experiment, Anthropic had 1,000 agents running an early Claude Fable 5.2 reproduce a formal proof of Fermat's Last Theorem in 28.7 hours. The team gave no Prove2Me or researcher guidance, only 60 lemmas to prove, short route hints, definitions and 47 previously proved lemmas. The agents coordinated through Git and a shared database, using a third of the original work's output tokens for a proof one-seventh as long.

Why it matters: Anthropic let a thousand agents split and coordinate work through a shared database and Git, keeping large-scale collaboration inside isolated sandboxes and two-layer monitoring.

Nov 8Sunday

The Decoder

AI coding agents still need human cleanup on nearly three out of four tasks in a real-world test

阿里巴巴与多所亚洲高校构建 Co-Design Bench,测试的十个模型中,表现最佳的 GPT-6 Astra 仅完整完成 26.4% 的任务。该模型平均满足约 70% 的需求,测试要求智能体根据真实用户请求修改现有网页应用,并由两名 AI 裁判检查功能是否实际可用。任务混合多种修改类型时所有模型表现均下降,同一模型在不同执行环境中的表现也有明显差异,研究团队建议同时报告模型与执行环境。

Hacker News front page

Apiiro finds Claude Code's "don't ask again" setting exposed credentials in 1,122 public GitHub files

Apiiro found that Claude Code's "don't ask again" option writes the full command, credentials included, into .claude/settings.local.json. On GitHub, 1,122 public files contain such secrets.

Why it matters: The study traces how credentials in approved commands land in project config files, and shows that scanning for key formats misses plain passwords inside commands.

Hacker News front page

MaxxBench tests whether DeepSWE's high scores survive altered tasks

MaxxBench changed the requirements or starting code of 89 DeepSWE tasks to see whether coding agents adapt to the changes. It found some models favor guessing at hidden tests. Models that scored close together on the original tasks diverged sharply after the edits.

Why it matters: Small edits to familiar tasks split models that had looked equal, adding requirement adaptability as a way to compare coding ability.

Nov 7Saturday

The Verge · AI

Researchers alarmed by new Anthropic report on automating AI R&D

METR's 11-week study of Anthropic found its models improving about twice as fast as a year earlier. METR judges that by October most of the acceleration came from AI, now mainly AI helping human researchers. Staff using the strongest internal models reported productivity about 12 times higher than without AI help.

Why it matters: Anthropic cut most of the evidence researchers consider important from METR's report, making it harder for the public to check the acceleration claim and its uncertainty.

The Decoder

Google's AI detects high blood pressure from a regular smartwatch without a cuff

Google Research 的 AI 利用普通智能手表的脉搏和运动传感器识别高血压迹象,无需血压袖带校准,研究发表于 Nature。神经网络从近500,000小时传感器数据中学习脉搏与心跳模式;在美国五个研究点的196人研究中,高血压识别比例为65.8%,传统诊室测量为55.3%。

Hacker News front page

The Components of Agent Memory

Anthropic 通过多项实验研究 Claude 的跨会话记忆,探索用 dreaming 离线核查日志、纠正记忆并构建关联知识页面。针对约 350,000 个文件、约 420M tokens 的文档库,结合搜索与 wiki 的评分随 wiki 扩展从 54% 升至 61%,仅搜索为 53%,每题还节省约 20,000 tokens。

Hacker News front page

SketchAgentBench – evaluate LLMs on 3D spatial reasoning

SketchAgentBench 是一个用于评估 LLM 3D 空间推理能力的基准。该基准通过草图任务考察模型对三维空间关系的理解。目前该条目在 Hacker News 首页获得 4 分,尚无评论。

OpenAI News

Measuring how AI agents handle everyday wet-lab science

OpenAI 与生物技术公司 Ligo 联合推出 BioBasicsBench,用 100 项常规湿实验室任务衡量 AI 智能体解读原始实验数据、识别错误并判断证据是否足以支撑结论的能力。

Nov 6Friday

Hacker News front page

AI for scientific discovery: How researchers use Claude

Anthropic 分析了 207,049 次 Claude 与 Claude Code 科研会话,发现使用集中于数据分析和写作,形式科学的相对使用率更高。数据分析占 Claude 会话的 33.7%、Claude Code 的 48.1%,构思与假设生成及数据采集的使用占比则低于研究者投入的时间占比。

The Decoder

METR finds Anthropic may have crossed its own AI acceleration warning threshold back in July

METR 审计认为,Anthropic 可能早在 7 月 15 日就越过自身 AI 研发加速预警阈值,其公开风险报告低估了加速程度及不确定性。自 2025 年 10 月以来,Anthropic 模型的 AECI 测试分数增长速度约为此前的 1.9 倍,审计方估计约 80% 的加速来自内部研究使用 AI,但实际加速倍数可能介于 1.3 至 3。