全部未入选
anthropics/claude-code released v2.1.309
ashwin-ant · 10-22 22:18 UTC
GitHub 来源显示，anthropics/claude-code 发布了 v2.1.309 版本。
AI 判断：标题初筛（未依据正文）：仅凭标题看，这是开发者工具的常规版本更新，未提供面向普通人的功能或可演示内容，属于技术类选题。
展开原始摘录
## What's changed
- Fixed sandboxed Bash commands on macOS and Linux running the shell's startup files outside the sandbox before each command
- Fixed macOS sandbox `denyWrite` glob patterns not applying at a path's real location when the pattern's folder is reached through a symlink
- Fixed sandboxed commands and auto-accepted edits being able to modify the MCP config file passed with `--mcp-config`
- Fixed a disabled MCP server being reconnected by a later permission prompt in print and SDK sessions
- Fixed the Bash permission check trusting a variable set by a command whose redirect can fail to open
- Fixed Bash permission checks not prompting for some compound commands that change directory before running git when the `cd` is wrapped or prefixed, as they do for a plain `cd`
- Fixed an `Edit(path)` deny rule not blocking a Bash redirect to that path in some commands that also held a subshell
- Fixed worktree-isolated sessions accepting some locale-prefixed commands the isolation guard should refuse
- Fixed worktree-isolated sessions on zsh accepting some arithmetic commands that the isolation guard should refuse
- Fixed worktrees and branch checkouts that Claude Code creates so they ignore git replace refs, as its other git operations already do
- Fixed profile store write protection not applying when the config directory is reached through a symlink
- Fixed sessions compacting repeatedly after reading files under a very large nested CLAUDE.md; a nested CLAUDE.md over the size limit now loads only its start
- Fixed ctrl+enter (send now) only queuing a message when pressed before the response had started, and restarting a prompt's hooks while leaving the message queued when pressed while those hooks were running
- Fixed key-bound slash commands running in the shell from `!` mode
- Fixed ctrl+o not showing messages from other agents and sessions that are still waiting to be delivered; they now appear in full under the transcript, marked as waiting
- Fixed long tool output previews being cut through an emoji, leaving session transcripts that `jq` could not parse
- Fixed Read reporting "File does not exist" for PDF page ranges in Linux sandboxes that share the parent's `/proc`
- Fixed messages from a Claude Desktop-hosted session sometimes being lost when its connection refreshed
- Fixed features missing from Claude Code sessions in the Claude desktop app when the terminal on the same computer was signed in to a different account or organization
- Fixed the `claude remote-control` status line showing `HEAD` as the branch for a folder that is not in a git repository
- Fixed split-pane agent teammates not inheriting the tool deny and `--tools` rules given on the leader's command line
- Fixed sessions renamed with `/rename` appearing to other sessions under a generated name after a VS Code window reload or a non-interactive `--resume`
- Fixed the `/rewind` list showing "No code changes" for a turn that changed checkpointed files through `/code-review --fix`, a subagent or a shell command
- Fixed `/privacy-settings` showing an out-of-date "Help improve our AI models" value; it now also says when the setting can't be fetched
- Fixed feedback saying "try again later" when your organization does not accept feedback; it now shows the server's reason and where to report a bug
- Fixed oversized /feedback reports failing instead of being resent without the transcript
- Fixed a skill in an `--add-dir` folder being unreachable when another skill already held its name; it is now offered as `
:
`
- Fixed `/verify` and `/code-review` not finding a nested package's own skill when its files were changed from the shell
- Fixed the /claude-api skill saying that `ant beta:worker poll --on-work` passes trace context to the command
- Fixed the /claude-api skill's C# tracing setup missing spans from the cloud platform clients
- Fixed the /claude-api skill saying that API spans for requests to Claude Platform on AWS can't join the caller's trace
- Fixed hooks that run in PowerShell not blocking when the hook's program exits with code 2, and on Windows `PreToolUse` and `PermissionRequest` hooks with `"shell": "powershell"` failing open when Git for Windows isn't installed
- Fixed a mod's `turn.step` hook that names a model by alias, such as `haiku`, failing every request with a model error: the alias now resolves to its model
- Fixed plugins and skills synced from claude.ai being removed, sometimes mid-session, after a disk read error
- Fixed artifact publishes and updates being refused for some ordinary pages, such as pages with certain minified scripts or artifacts Claude could not read first
- Fixed Claude asking for permission on each artifact update for the rest of a turn after the conversation was compacted in that turn
- Fixed fine-grained tool streaming staying off on Amazon Bedrock when `ANTHROPIC_BEDROCK_BASE_URL` is set to the default `bedrock-runtime.
.amazonaws.com` endpoint
- Fixed the "Waiting for API response" hint flickering behind gateways that send only keepalive pings: routes off Anthropic's backend now wait 45 seconds
- Fixed Claude notifying unrelated GitHub users by writing role names and placeholders as mentions in PRs, issues and comments
- Linux: Fixed sandbox write denies lifting when a background command outlived the session
- Windows: Fixed PowerShell permission checks auto-approving some `Set-Location` or `cd` commands that target the current directory and carry extra arguments; these now prompt for approval
- macOS: Fixed the "Install Command Line Developer Tools" dialog opening each time Claude Code starts on a Mac without Xcode or the Command Line Tools
- Self-hosted runner: Fixed `--startup-timeout-min` stopping a session whose first prompt, such as `/code-review`, was still working in a subagent; a frozen process is still stopped
- Improved the speed of many git commands Claude Code runs for itself on Windows; on every platform, errors from those commands now show in English even when git is set to another language
- Improved the error when a hooks module is nested too deep to load: it now names the file instead of showing a bare stack-overflow message
- Improved disk-full handling: a Bash command cut off by a full disk now says so and names the folder to free up, and starting on a full temp volume prints a clear message, not a raw ENOSPC error
- Improved the /claude-api skill to offer API tracing alongside SDK tracing when asked to add OpenTelemetry
- Improved the /claude-api skill to check the live docs before saying a feature is unsupported
- Self-hosted runner: Improved sessions that move to a new runner: Claude now sees the tool call that was interrupted, marked as outcome unknown
- Updated the bundled dataviz skill to accept APCA as a text-contrast measure (keeping WCAG 2 ratios where AA conformance is required), add an `apca(text, background)` function to its palette validator scripts, and say whether a label inside a colored chart fill should be white or near-black
- Updated the /claude-api skill: service account keys and Workload Identity Federation tokens can now manage OpenTelemetry destinations
- Changed `sed -i` to get the standard Bash prompt and run as written, not a simulated file-edit preview; like other Bash edits, it is no longer restored by `/rewind`
- Changed `claude ultrareview` to require workspace trust: it asks on a terminal, else refuses; scripts pass `--trust-workspace`, which takes no value (put a PR number or base branch before it), or set `CLAUDE_CODE_DISABLE_ULTRAREVIEW_TRUST_CHECK=1`
- Changed the dangerous `rm` and `rmdir` prompt in bypass permissions mode in the Desktop app and VS Code to deny the command at once with a rewrite hint, so unattended sessions keep going
- Changed commands the settings exclude from the sandbox to ask before writing to its write-list paths while the sandbox is enabled but cannot run; the unsandboxed fallback is refused there
- Changed worktree creation and reuse so git no longer fetches objects on demand, in ordinary clones and in partial clones with a Bash sandbox configured; such a partial clone fetches them first
- Changed the `/claude-api` skill to build Managed Agents subagent teams on `multiagent_20261001` with dynamic workflows on by default, instead of `coordinator`
- [VSCode] Added a log warning that says what to do when dictation or the extension's own requests fail certificate verification, such as behind a TLS-inspecting proxy
- [VSCode] Fixed the Claude panel staying blank after the editor cached an incomplete copy of its files; it now recovers by itself, or says what went wrong and how to fix it
- [VSCode] Fixed Session history filling up with empty "Untitled" sessions each time you start a new session without sending a message
- [VSCode] Fixed long slash-command messages such as `/goal` filling the chat panel; they now fold behind Show more like other prompts (anthropics/claude-code#97469)
- [VSCode] Fixed long messages not folding behind Show more after the chat panel is made narrower
- [VSCode] Fixed Ctrl+J in the chat input always inserting a newline: it is now left to VS Code and the OS when "ctrl+j" is set to null in the Chat context of ~/.claude/keybindings.json
- [VSCode] Fixed a mod's "From the … plugin" label above the prompt being drawn over the slash command and @-mention menus (anthropics/claude-code#102981)
- [VSCode] Fixed an empty list item, paragraph or code block, and list items renumbered from 1, when pasting copied LaTeX math
- [VSCode] Fixed the feedback error showing only an HTTP status when your organization does not accept feedback
- [VSCode] Improved Edit and Write speed in the chat panel: the wait for new problems after each edit is now 0.75 seconds in most cases, down from 1 to 1.5 seconds
- [Cloud sessions] Fixed the Add connector list in routine forms saying no connectors were available while they were still loading
- [Cloud sessions] Fixed the cloud environment editor shuffling environment variables and adding backslashes to values that contain both kinds of quote
- [Cloud sessions] Fixed the self-hosted environment details dialog in admin settings so its title, tabs and buttons stay in view while only the content scrolls
- [Cloud sessions] Fixed the "read documentation" tool row in cloud sessions, which showed escaped JSON instead of a readable page
- [Cloud sessions] Improved a cloud session's Changes pane, which now says when it lists only pushed commits or only Claude's edits
- [Claude Tag] Fixed Claude in Slack posting a draft you approved for a channel into your DM by mistake; a post with no channel is now refused so it can go to the approved channel
- [Claude Tag] Fixed Allow and Deny on Claude's approval cards in Slack doing nothing when the session had stopped; your answer now restarts the session
- [Claude Tag] Fixed Claude ignoring untagged messages in a Slack channel after its workspace moved to a different Claude organization, including unshared channels on Enterprise Grid
- [Claude Tag] Fixed Claude in a Slack channel posting a threaded reply before reading a newer message someone had just added to that thread
- [Claude Tag] Fixed the Slack page in Claude Tag admin settings saying no workspaces were connected when the list failed to load; it now shows an error with a Retry button
- [Code Review] Fixed a finished review showing no check on the pull request when earlier tries to create it failed; a final try now adds a completed check
- [Code Review] Fixed a push queued behind a running review still starting a review after `@claude review stop`; its check now closes and says why
Jared Kushner Hacked Phone Reveals Secret Negotiations, Millions in Transactions
ck2 · 10-22 21:24 UTC
报道称库什纳的手机被黑客入侵，泄露了秘密谈判内容和数百万美元的交易信息。
AI 判断：标题初筛（未依据正文）：仅凭标题看，这是政治人物的手机入侵新闻，与AI无关，普通人既无法亲自验证也无法实际使用，不适合作为AI选题。
展开原始摘录
--- Top Comments ---
[consumer451]: I am pretty darn jaded these days, but I find myself staggered. edit after reply: > What was odd, the source said, was that the money in the account—hundreds of millions of dollars—had come from a bank in Azerbaijan. Ahh, now the arrest of the former chief of Germany's BND, related to what seemed like a random country "Azerbaijan" is making a bit more sense?
[cf100clunk]: http://archive.today/FHGAt
[enjoyyourlife]: I'm more interested to see if it revealed he is using it on Saturdays
[Calavar]: So Kushner's phone was hacked around the same time as Hunter Biden's laptop? Why has the latter captured the media so much more then the former?
[ck2]: it's going to take decades to unravel all the corruption by this administration it is staggering and they are destroying records in violation of law because "pardons for everyone" on the way out
Replit：Blog（RSS） · 10-22 20:24 UTC
Replit 发布 Monitoring Agent，这是一个面向应用、可自动修复并部署的值班智能体。
AI 判断：标题初筛（未依据正文）：这是面向开发者的运维智能体，标题未体现普通用户可直接使用的场景，按技术类处理。
展开原始摘录
Replit 发布 Monitoring Agent，为应用提供相当于自主值班工程师的监控智能体。它可以读取生产日志和指标调查错误、变慢和部署失败，在启用 autofix 后验证修复并邮件汇报，启用 auto-publish 后可自动部署修复。该能力建立在早前上线的 App Monitoring 之上，用户可在应用的 Publishing 设置中开启。
Claude 网页、桌面和移动端将默认启用 auto mode
X：Claude (@claudeai) · 10-22 20:04 UTC
Claude 将在网页、桌面和移动端默认启用 auto 模式。
AI 判断：标题初筛（未依据正文）：标题显示 Claude 多端将默认开启 auto 模式，属于用户可感知的产品变化，但细节不足，难以做成可验证教程。
展开原始摘录
Claude 官方宣布，下周起 Pro、Max 和 Team 套餐在 Claude 网页、桌面和移动端应用中的会话将默认以 auto mode 运行。auto mode 不再逐步请求用户批准，而是由自动检查拦截高风险操作，用户可随时切回手动模式。其引用的 Claude Code 更新提到 auto mode 用独立分类器审查 shell 命令，测试中拦截了 89% 的危险命令，手动批准为 14%。
OpenAI Launches Decisions API for 10x Faster App Decisions Using GPT-6 Luna
Future Tools · 10-22 20:00 UTC
标题称 OpenAI 发布 Decisions API，宣称借助 GPT-6 Luna 让应用内的决策速度提升 10 倍。
AI 判断：标题初筛（未依据正文）：仅从标题看，这是面向开发者的 API 发布，且“GPT-6 Luna”与“10 倍”缺少来源佐证，普通用户无法直接使用，也无法演示。
展开原始摘录
Tools & Products · 3h ago OpenAI Launches Decisions API for 10x Faster App Decisions Using GPT-6 Luna OpenAI has made its Decisions API generally available, built on GPT-6 Luna to handle small, frequent application decisions up to 10x faster than GPT-6 Luna via the standard Responses API. Developers define a question and allowed answers, provide context as text, images, or native JSON, and receive a probability, choice, or score. New features include native GPT-Live integration, prompt caching, US and EU inference residency, and support for up to 400 questions per request. Pricing is $0.10 per million input tokens, with no output-token charges.
The Verge：AI（RSS） · 10-22 19:45 UTC
据标题，The Verge 分析 SpaceX 与特斯拉设立特别委员会、讨论两者合并背后的财务逻辑。
AI 判断：标题初筛（未依据正文）：仅从标题判断，这是企业合并与财务层面的新闻，与普通人可用的 AI 功能或生活场景无关，也缺少可演示的画面。
展开原始摘录
The Verge 作者 Elizabeth Lopatto 分析 SpaceX 和 Tesla 各自成立特别委员会讨论合并一事。
DE STAAT - FREEDOM (Official Music Video)
Simon Willison · 10-22 19:42 UTC
仅从标题看，这是一条名为《FREEDOM》的官方音乐视频，与 AI 无关。
AI 判断：标题初筛（未依据正文）：仅凭标题判断，这是音乐视频内容，不含任何 AI 新功能、工具或可演示的实用信息，不适合目标账号。
展开原始摘录
DE STAAT - FREEDOM (Official Music Video)
OK well this is a first for me... at
00:44
in this new music video by Dutch alternative rock band
De Staat
:
The visual effects are credited to Matties Grooten of
Fake it or Leave it
.
Via
flir on Hacker News
Tags:
music
,
ai
,
generative-ai
,
pelican-riding-a-bicycle
Microsoft's MAI-DxO AI Outperforms Physicians in Diagnostic Accuracy at Half the Cost
Future Tools · 10-22 19:00 UTC
微软 MAI-DxO AI 据称在诊断准确率上超过医生，并将成本降低一半。
AI 判断：标题初筛（未依据正文）：标题仅给出诊断准确率与成本对比，缺少普通人可用的产品、地区开放和验证信息，偏技术研究。
展开原始摘录
Research · 4h ago Microsoft's MAI-DxO AI Outperforms Physicians in Diagnostic Accuracy at Half the Cost Microsoft's MAI-DxO AI diagnostic system outperforms physicians in accuracy at lower cost, according to research published in Science. On complex New England Journal of Medicine cases, MAI-DxO achieved 83% diagnostic accuracy versus 36% for physicians using reference resources. On typical emergency department cases from MIMIC-IV-ED, it reached 80% accuracy versus 47% for physicians. The system operates like a panel of virtual physicians debating diagnoses across multiple rounds. It achieves results at half or one-third the cost, and Microsoft plans real-world testing at Beth Israel Deaconess Medical Center.
OpenAI News · 10-22 19:00 UTC · 同链接合并评分
标题称 OpenAI 发布 Decisions API，宣称借助 GPT-6 Luna 让应用内的决策速度提升 10 倍。
AI 判断：标题初筛（未依据正文）：仅从标题看，这是面向开发者的 API 发布，且“GPT-6 Luna”与“10 倍”缺少来源佐证，普通用户无法直接使用，也无法演示。
展开原始摘录
Fast, focused, low-cost decisions for any application.
OpenAI：官网动态（RSS · 排除企业/客户案例） · 10-22 19:00 UTC · 同链接合并评分
标题称 OpenAI 发布 Decisions API，宣称借助 GPT-6 Luna 让应用内的决策速度提升 10 倍。
AI 判断：标题初筛（未依据正文）：仅从标题看，这是面向开发者的 API 发布，且“GPT-6 Luna”与“10 倍”缺少来源佐证，普通用户无法直接使用，也无法演示。
展开原始摘录
OpenAI 宣布基于 GPT-6 Luna 的 Decisions API 正式可用，用于分类、路由、评分等有界答案问题，速度最高比通过 Responses API 使用 GPT-6 Luna 快 10x。
Donating Agent Skills to the Agentic AI Foundation
Simon Willison · 10-22 18:09 UTC
该文标题显示，作者将 Agent Skills 捐赠给 Agentic AI Foundation。
AI 判断：标题初筛（未依据正文）：标题指向开发者代理技能生态与基金会治理，普通人难以直接使用或演示，属技术类。
展开原始摘录
Donating Agent Skills to the Agentic AI Foundation
News today from Anthropic:
We’re committed to keeping Agent Skills open, community-driven, and vendor-neutral. Today, we are donating the standard to the Agentic AI Foundation (AAIF) under the Linux Foundation,
which we co-founded
with Block and OpenAI last year, and to which we donated the Model Context Protocol (MCP).
I'm pleased to note I called this one
last December
, when Anthropic first declared skills to be an "open standard":
I wouldn't be surprised to see this end up
in the AAIF
, recently the new home of the MCP specification.
Fun side note: since my blog content is available to Claude via the
datasette-mcp
plugin I found that post by
prompting Claude
to "
Use simonwillison.net MCP to find if I ever talked about how I want skills to be donated to a standards body
".
Tags:
standards
,
ai
,
generative-ai
,
llms
,
anthropic
,
model-context-protocol
,
skills
Microsoft AI：官方博客（网页） · 10-22 18:01 UTC · 同链接合并评分
微软 MAI-DxO AI 据称在诊断准确率上超过医生，并将成本降低一半。
AI 判断：标题初筛（未依据正文）：标题仅给出诊断准确率与成本对比，缺少普通人可用的产品、地区开放和验证信息，偏技术研究。
展开原始摘录
Microsoft AI 发布 Sequential Diagnosis Benchmark（SDBench）和 AI 诊断系统 MAI-DxO，论文发表于 Science。
Anthropic Donates Agent Skills Standard to Agentic AI Foundation Under Linux Foundation
Future Tools · 10-22 18:00 UTC
据标题，Anthropic 把 Agent Skills 标准捐给 Linux 基金会旗下的 Agentic AI 基金会。
AI 判断：标题初筛（未依据正文）：标题显示这是面向开发者的智能体标准与生态治理新闻，普通用户难以直接使用或演示。
展开原始摘录
Company News · 5h ago Anthropic Donates Agent Skills Standard to Agentic AI Foundation Under Linux Foundation Anthropic has donated Agent Skills, its open standard for giving AI agents specialized knowledge and workflows, to the Agentic AI Foundation (AAIF) under the Linux Foundation. Originally developed while building Claude Code and published as an open standard in December 2025, Agent Skills has been integrated by over 70 platforms including ChatGPT, Cursor, GitHub Copilot, and VS Code. One million skills have been published and nearly 280 million installed. Engineers from Anthropic, Google, and OpenAI will maintain the standard, with David Soria Parra and Mahesh Murag as Lead Maintainers.
OpenShell – A safe, private runtime for fleets of autonomous AI agents
niemandhier · 10-22 17:44 UTC
英伟达开源 OpenShell，一个为大量自主 AI 智能体提供安全、私有运行环境的运行时。
AI 判断：标题初筛（未依据正文）：标题表明这是面向开发者的智能体底层运行时设施，普通人无法直接使用或演示，属于纯技术选题。
展开原始摘录
--- Top Comments ---
[fennecbutt]: I've been using iron proxy recently for a project I just started, it does network restrictions by domain etc with plenty of options, credential swapping and more. I have no affiliation with it, but it solved most of what I need in one go when I was stressing out about it. Only real problem remaining if you truly want to be able to trust agents is how to prevent data (in my case codebases the agent works on) exfiltration a la HF incident. We can restrict network to trusted domains, but th...
[crypt1d]: I tried to vibecode something similar last weekend. Got tripped up by Claude's safety classifier. It wouldn't let me build any features that track/audit process executions , not through custom kernel hooks nor even using off the shelf software for it.
[mfro]: If you haven’t already been running your agents in containers, you’re crazy. That said, this looks like a great way to manage those containers.
[smallnix]: > Agents are most useful when they can read files, install packages, call APIs, and use credentials. OpenShell gives them that capability without giving them unrestricted access to your data, secrets, or network. To me this obfuscates an oxymoron: the more access you give the agent the more capable it becomes. You can't have you cake and eat it, this tool allows to strike a balance I guess. But the blurp reads like it can do it all, but now secure. That can't be true.
[StarlaAtNight]: What are people actually doing with fleets of agents?
A farewell to Arc browser
suhdonghwi · 10-22 16:27 UTC
这是 Arc 浏览器的告别消息。
AI 判断：标题初筛（未依据正文）：标题仅显示 Arc 浏览器告别，与 AI 无直接关系且信息有限。
展开原始摘录
--- Top Comments ---
[jamie_ca]: Writing was on the wall once they pivoted to Dia, I'm pretty happy having switched to Zen (though for a few things that matter, it's Firefox-based not Chrome). https://zen-browser.app/ They picked up a lot of cues from Arc - sidebar tabs, pinned vs generic tabs, global bookmarks, spaces...
[Aurornis]: I had to look Arc up because I forgot what it was. For others confused like me: Arc was an alternative web browser with novel UX and AI features built-in. The company was acquired for $610 million by Atlassian and development was halted in favorite of Dia, another browser with AI features and novel UX (which I also had to look up)
[throwaway13337]: Arc had a lot right. The closest thing right now is Orion. It does the vertical tabs and the page modification in the same way. Webkit based. It's made by the Kagi guys who I really agree with - bootstrapped and not VC backed. A sustainable business. And charging for a service instead of making your customers the product. That's real ethics to me. Their browser is not open source, though, which is a major blow. But it would seem most browsers are not truly now, if you look into it. ...
[constantlm]: I've been expecting this news for a while, and I moved to Zen when they changed direction originally. Can highly recommend it as a near drop-in replacement for Arc, albeit built on Firefox. https://zen-browser.app/
[Zambyte]: I was expecting this to be about http://arclanguage.org/
Getting ready to operate GNU Taler in Germany
dewey · 10-22 16:14 UTC
GNU Taler 宣布正准备在德国投入运营。
AI 判断：标题初筛（未依据正文）：标题仅涉及 GNU Taler 在德国的运营准备，与面向普通人的 AI 工具或应用无关。
展开原始摘录
--- Top Comments ---
[Eueudhsbsj32]: I don't see how EU regulators would allow payment privacy in any meaningful way. It will be choked out at key junctions with KYC in the same way crypto has.
[F3nd0]: Wishing Taler lots of success and hoping for the best. Adopting something like this should be a given, but unfortunately freedom and privacy aren’t seen as essentials, not even in the EU.
[Deukhoofd]: So this is trying to compete against Wero, which is being backed by most major western European banks? I wish them luck, but I somewhat doubt they can compete against that.
[jttnr]: I'm a happy GLS Bank customer, and I've just learned about its involvement in Taler. Pretty awesome. Europe desperately needs its own digital payment solutions to reduce its dependence on US companies and its exposure to sanctions that could target EU citizens if the president has a bad day. GLS is also on board with Wero, which, unfortunately, hasn't seen big adoption yet. I really hope Taler takes off.
[hangrybear666]: Never having heard of it before, the moment I read "cash like tokens in your account" I was immediately out. Also claiming buyer privacy when the money ultimately has to come from a bank transaction or another digital transaction leaving some sort of trail is a false illusion.
Every AI model has the same "controversial" take on humans
rakannimer · 10-22 16:08 UTC
一篇文章指出，不同 AI 模型在涉及人类的争议性问题上给出了相同看法。
AI 判断：标题初筛（未依据正文）：标题仅暗示模型观点趋同，缺乏具体案例和可演示性，对普通人的实用价值有限。
展开原始摘录
--- Top Comments ---
[skyzoidbroczky]: What's the point of asking that in the first place, people still do not understand in big 2026 that's it's just a probabilistic database, it has no opinion, it does not know, the only thing it does is reproduce patterns more or less abstract in function of the pressure you put on certain representation during training time.
[keerthiko]: Ah, the irony that all the models produced a shared opinion while pretending they came up with it from analysis, which from their statements is implicitly some kind of a "human flaw" is not lost on me.
[kens]: I took a quick look at the data, and it doesn't support the conclusions. The thesis is that AI models have the take: "We pick what our group believes, then invent the reasons." However, GPT-6 Astra said, "Humanity’s biggest moral failure is the talent for making convenience look like principle." That statement has nothing to do with group beliefs or conformity. At best, you could count it as a half-match, since it deals with invented reasons.
[daveloyall]: I thought the article was going to be about system prompts, or model specs, or "constitution documents". Isn't it? Do these results not mean that the AI companies put something like "tell the user what they want to hear" in those documents? (Or whatever phrase would produce those results.)
[a13o]: Train your model on Very Online corpuses, ask Very Online questions, get Very Online answers
PHP vulnerabilities affect 70% of web, AI-discovered exploits hit Wikipedia, WordPress, governments
Future Tools · 10-22 16:00 UTC
据报道，PHP漏洞影响约70%的网站，AI发现的漏洞利用已波及维基百科、WordPress及政府机构。
AI 判断：标题初筛（未依据正文）：仅凭标题判断，这是偏技术的安全新闻，对普通人的直接实用性较低。
展开原始摘录
Policy & Governance · 7h ago PHP vulnerabilities affect 70% of web, AI-discovered exploits hit Wikipedia, WordPress, governments Two critical PHP vulnerabilities, CVE-2026-109618 and CVE-2026-109619, are disrupting roughly 70 percent of the web after being discovered using GPT-5.6 Sol by a researcher at Calif. The flaws, rated 9.2 out of 10 in severity, affect every PHP version dating back 20 years and enable remote code execution. Wikipedia, WordPress, 4chan, DeviantArt, the Internet Archive, and Dutch and Finnish government sites were taken offline. Cloudflare logged millions of exploit attempts within hours of public disclosure, and some older sites face permanent shutdown.
Anthropic 股权表内幕：风投迟到多年，如今无人持股超 2%
Newcomer 新闻长文（RSS） · 10-22 15:53 UTC
一篇报道梳理 Anthropic 的股权结构，指出风投机构较晚才入场，目前没有任何单一股东持股超过 2%。
AI 判断：标题初筛（未依据正文）：标题显示这是融资与股权结构报道，缺少普通人可用的 AI 功能或可演示内容，按选题偏好应低分。
展开原始摘录
Newcomer 长文复盘 Anthropic 从 OpenAI 出走团队到即将 IPO 的股权演变。据投资人文件和多位知情人士，多年来风投因公司治理结构和安全优先路线而观望，目前没有任何风投机构持股超过 2%，Anthropic 曾向投资人明确将单一投资人持股控制在该水平以避免话语权集中。
Quoting Armin Ronacher
Simon Willison · 10-22 15:34 UTC
该条目仅标明是对开发者 Armin Ronacher 话语的引用，标题本身没有说明具体话题或内容。
AI 判断：标题初筛（未依据正文）：标题只有一句引用、没有可核实细节或可演示功能，仅凭标题无法判断对普通人的价值，暂按技术类低分处理。
展开原始摘录
As fun as AI is, we built so much bad software over the last 30 years, and now the universe is punishing us for our sins.
—
Armin Ronacher
Tags:
security
,
generative-ai
,
armin-ronacher
,
ai
,
ai-security-research
,
llms
METR 与 Redwood Research 红队测试 OpenAI 内部智能体监控系统，发现多个漏洞
Redwood Research：Blog（RSS） · 10-22 15:32 UTC
该文称 METR 与 Redwood Research 红队测试了 OpenAI 内部智能体监控系统，并发现多个漏洞。
AI 判断：标题初筛（未依据正文）：仅据标题判断，这是智能体监控与红队安全研究，偏技术，普通用户难以直接使用或演示。
展开原始摘录
METR 与 Redwood Research 各派一名员工，于 7 月 21 日至 8 月 11 日对 OpenAI 内部失准监测系统开展三周高权限试点压力测试，产出 61 页内部报告。
Run full iOS 16.5 firmware on Android Devices
aghuang · 10-22 15:26 UTC
标题称可在 Android 设备上运行完整的 iOS 16.5 固件。
AI 判断：标题初筛（未依据正文）：这是基于标题判断的内核移植类技术项目，并非普通人可直接使用的 AI 工具，也缺少可验证的日常场景。
展开原始摘录
https://hackios-web.vercel.app/en
--- Top Comments ---
[aghuang]: I found the website to be quite nauseating to look at, so I decided to use the GitHub link instead. But it's quite impressive that it works on QEMU, POCO X5 Pro 5G and some Xiaomi devices too. Website is here: https://hackios-web.vercel.app/en Source Code: https://github.com/oosexgang/hackios-xnu-kernel
[SoKamil]: > It is a research bench, not a ROM: a session lasts a given number of seconds, then the kernel returns to the loader and hands over its log. I wonder if it is here because Claude would deploy safeguards and not proceed with the project otherwise.
[drnick1]: It's interesting that this can be done, but why would anyone want to run iOS on Android devices? A more desirable outcome would be running a free Android distro like Graphene on Apple hardware. Or indeed just liberating phones controlled by Google by hacking the bootloader and replacing the OS.
[sanid]: This is going to be fun. There are already replica phones for the flagship phones that are running a shitty android version that has been customized to look like iOS. People need to tripple check now when buying older used iPhone models.
[pmkary]: Someone finally did this, wow! 2026 just keeps getting more crazy.
PHP2Shell 漏洞披露前黑客已有四天窗口，多国机构紧急响应
Ars Technica：AI（RSS） · 10-22 15:25 UTC
Ars Technica 报道称，PHP2Shell 漏洞在公开披露前已被黑客利用四天，多国机构正在紧急响应。
AI 判断：标题初筛（未依据正文）：这是网络安全事件，与 AI 及普通人的实用场景关联较弱；仅凭标题也无法核实漏洞细节，因此选题价值有限。
展开原始摘录
PHP2Shell 漏洞在被披露前四天已遭黑客利用，Calif 与 Wiz 的研究人员发现这些漏洞。OpenAI 称其前沿 AI 在攻击者之前发现了漏洞，并与 PHP、Calif、Wiz 合作验证并开发补丁。NSA 建议无法立即更新的机构将系统离线，英国 NCSC 表示漏洞可能让黑客无需登录凭据即可控制网站服务器，白宫称联邦机构正在评估对政府系统和关键基础设施的潜在影响。
Quad9 (DNS Resolver) to Exit France
farrokhi · 10-22 14:32 UTC
据标题信息，DNS 解析服务 Quad9 发布博客宣布将退出法国市场。
AI 判断：标题初筛（未依据正文）：这是与隐私和网络基础设施相关的技术新闻，与 AI 及普通人的日常使用关系较弱，仅凭标题也无法核实细节。
展开原始摘录
--- Top Comments ---
[johnhtodd]: (context: I'm the CTO of Quad9 and am one of the people dealing with the lawyers) There is an understandable focus on "how are you going to block French users" when we have also claimed that geoIP blocking is insufficient for selectively denying resolution for certain names - we understand this conflict, but we think it is energy spent in the wrong place. It is probable that we will use geoIP filters to block French users from our services after November 24. We could apply thi...
[nadermx]: "The plaintiff (Sony Music Entertainment GmbH) then used a VPN to access the blocked content from Germany in circumvention of the geoIP blocking. The court agreed with Sony in their assertions of noncompliance, and we were fined for contempt of court." Imagine doing everything possible to stop something. They then circumvent your protections with yet another service, and you get fined. I wish I could even suggest what someone could do against these trolls. This is like saying "...
[euio757]: Seems they're following Cisco similar move 2y ago of exiting the market > Canal+ has been busy getting web-blocking injunctions in France against sports piracy sites which stream sporting events for which it has the exclusive broadcast rights within the French market. It previously targeted DNS resolvers operated by Google, Cloudflare and Cisco. > All three unsuccessfully opposed the web-blocking orders. Google and Cloudflare, therefore, have been forced to instigate them. However,...
[JokerDan]: DNS resolvers getting sued should stop resolving domains/properties of the companies suing them too when they are forced to take these kinds of actions/steps off the back of a legal decision like this.
[babarock]: Punishing the DNS resolver is weird. Would you punish the phone book company for having the number of a criminal listed?
404 Media（RSS） · 10-22 14:23 UTC
404 Media 分析 PHP2Shell 事件，讨论 AI 正在如何改变黑客攻防格局。
AI 判断：标题初筛（未依据正文）：该内容偏网络安全与攻防分析，主要面向开发者或安全从业者，普通用户难以直接使用，也缺少可演示画面。
展开原始摘录
404 Media 的 Jason Koebler 梳理近一个月安全专家对 AI 挖洞能力引发的"Black October"恐慌。Wiz、Calif 和 OpenAI 的研究者用 AI 发现了 PHP 的 PHP2Shell 漏洞，Cloudflare CEO 称公开公告前 10 小时已见数千次利用尝试，公告后 2 小时内增至数十万，目前达数百万级。
What’s getting pwned right now
mrkd · 10-22 14:08 UTC
标题意为展示当前正在被攻破或拿下的对象。
AI 判断：标题初筛（未依据正文）：仅凭标题看，这是模糊的实时攻破动态，未说明与普通人的 AI 用途，需核实内容。
展开原始摘录
--- Top Comments ---
[pixl97]: Would be interesting to see what 2025 looked like in comparison. The rate at which flaws are being found is just insane. Many people warned us in the past that our lax security postures would come back to bite us, and we're finally here.
[demibabs]: Every ai generated page tries to cram way too much information in way too small a place
[b1n]: The Sponsors layout looks like a list of the most hacked sites to someone not paying attention.
[xbar]: After the first 8 hours of playing with Mythos, I was asked what I thought. My response, "this generation of LLMs is going to drive the largest wave of security incidents the world has ever seen." It is still getting started.
[aeturnum]: I don't understand the categorization for incidents. This signal vulnerability[1] is listed as "Live" and "ongoing" but was patched over a month ago (just announced today). If an incident with no known exploits is "ongoing" 1.5 months after the patch is released, when would it be over? [1] https://pwn.watch/i/signal-ringrtc-cve-2026-113380
Sierra：Blog（RSS） · 10-22 14:02 UTC
Sierra 发布 𝜏-Entity 基准，用于检验语音智能体能否准确采集姓名、电话等实体信息。
AI 判断：标题初筛（未依据正文）：标题显示这是面向语音智能体实体采集能力的基准评测，偏技术评估，普通人的直接实用价值有限。
展开原始摘录
Sierra 发布 𝜏-Entity 基准，测量语音智能体能否正确采集、识别错误并修复姓名、电话号码、地址和邮箱等细节，包含 10 种实体类型共 200 个任务，并与 𝜏-i18n 一样覆盖六种语言。
eMudhra to exit global SSL/TLS certificate business
bo0tzz · 10-22 13:33 UTC
eMudhra 宣布将退出全球 SSL/TLS 证书业务。
AI 判断：标题初筛（未依据正文）：标题仅涉及证书机构业务退出，属于技术基础设施新闻，对普通人的直接相关性低。
展开原始摘录
--- Top Comments ---
[rdtsc]: > However, the level of social media amplification around this incident has brought scrutiny that is out of proportion to SSL/TLS’s place in our business Scrutiny is inversely proportional to bit complexity of the private keys they used. Without a failure to apologize or expose how it happened: a human mistake, process mistake, etc. this looks pretty bad on them. > eMudhra’s global SSL/TLS infrastructure runs separately from its other operations. The company’s Global digital s...
[Toutouxc]: Somehow this is making it even worse. Like, I'm having issues explaining the whole thing to my wife, because I can't come up with a metaphor or parable that's even remotely as ridiculous as what they've done. Anyone's got a good one? Maybe, like, setting up a factory that builds both LHD and RHD cars and building cars with two passenger seats in the front for two years straight without anyone noticing?
[ieie3366]: Related: the absolutely brutal question battery the Chrome team publicly submitted to eMudhra https://bugzilla.mozilla.org/show_bug.cgi?id=2083449#c30
[luispa]: -----BEGIN RSA PRIVATE KEY-----
4
-----END RSA PRIVATE KEY-----
[rahimnathwani]: "There were no end entity certificates issued out of the affected CA’s." This seems like a weird statement given that anyone can issue certificates out of the affected CAs.
Woman faces six years in prison over Cash App account she never opened
oidar · 10-22 13:11 UTC
一名女子因一个她从未开通的 Cash App 账户被关押三晚，并面临六年监禁。
AI 判断：标题初筛（未依据正文）：标题基于，这是身份盗用/诈骗新闻，与 AI 无关，不适合作为 AI 选题。
展开原始摘录
--- Top Comments ---
[benjiro29]: * authorities have offered Smith a plea deal including a $1,000 fine, $1,000 in restitution and three years' probation.* This is the goal ... They know they are at fault but admitting they made a mistake is damaging to careers. In a lot of cases a carrot is dangled in front of people ... You can wast a lot of time (and money) fighting for your innocent, and take the risk your still going to get convicted. Or you can take the plea deal, and it absolves us from our mistakes. And you save m...
[squidbeak]: > Meanwhile, authorities have offered Smith a plea deal including a $1,000 fine, $1,000 in restitution and three years' probation. She has refused, saying she is innocent. This is nuts. Is it possible US prosecutors give so few shits about justice?
[Roark66]: It will never stop until prosecutors/police become liable for gross negligence.
[jawiggins]: I recall once reading some Patrick Mckenzie article where he discussed how administrative issues like this used to be easily fixed by journalists. They could reach out to an institution seeking comment on a story, the institution's PR department could then ask for 48 hours to comment, in which time they could run something up the flag pole and get it fix. But with the speed on online reporting, they're now incentivized to report on the scandal ASAP, leading institutions to similarly...
[doodlebugging]: Time for CashApp founders and C-levels to be held personally liable for creating a system with so few checks that it appears to be designed to facilitate fraud and theft. Then you sue all the law enforcement officers and local officials who are attempting to lay the blame for this on someone they should know is likely innocent. Make all of the financial liability for this lazy investigation come directly from the pockets of the cops and prosecutors until they have nothing left to their names ...
Wiz：Research（安全研究） · 10-22 13:05 UTC
Wiz 研究称，威胁组织 Storm-0940 利用名为 ARTEX 的智能体扩大对 IT 与软件厂商的攻击。
AI 判断：标题初筛（未依据正文）：这是面向 IT 与安全团队的攻击情报，普通用户难以直接使用，且标题信息有限、缺少可演示画面。
展开原始摘录
Wiz 披露 2026 年 10 月有攻击者使用开源自主渗透测试框架 ARTEX 针对多家 IT 和软件厂商，活动与微软报告的中国国家级关联组织 Storm-0940 重叠，并已扩展到其他软件领域的供应链攻击尝试。
Claude’s personal agent
Ben's Bites · 10-22 13:01 UTC
该文标题提到 Claude 的个人智能体，具体内容尚不明确。
AI 判断：标题初筛（未依据正文）：仅凭标题无法确认普通人可用性、演示效果或使用限制，需查看正文后再判断实用价值。
展开原始摘录
And how I run the newsletter with agents
What is voice over? Definition, examples, and how it works
ElevenLabs Blog · 10-22 12:00 UTC
介绍配音的定义、示例以及它是如何运作的。
AI 判断：标题初筛（未依据正文）：标题显示这是配音概念科普，未体现新的AI功能或时效性事件，实用性和新奇度有限。
展开原始摘录
Learn what is voice over, how it differs from voice acting and dubbing, and how to write a script and create a voice over with AI in minutes.
Xbox wants you to pay 24 to 40 cents per hour to stream games you bought
entryleveljobsc · 10-22 11:42 UTC
据标题，Xbox 计划对玩家串流自己已购买的游戏按每小时 24 至 40 美分收费。
AI 判断：标题初筛（未依据正文）：这是游戏云串流的收费新闻，与 AI 无关，也缺少普通人可用的 AI 演示或实用价值（仅据标题判断）。
展开原始摘录
--- Top Comments ---
[paxy]: Why do they keep stressing the "games you already bought!!!" part? There is no charge to play the game on your own hardware. This pricing is for renting the GPUs, same as GeForce Now and plenty of other services. And it's a pretty solid deal for people who don't want to build a multi-thousand dollar rig for gaming 10-20 hours a week.
[ilikeitdark13]: To be honest, this actually seems a bit understandable Unlike many things in gaming these days.
[threepts]: Seems reasonable for your average user who games 10-12 hours a week and would not want to go through the hassle of owning and maintain a physical console (plus paying electricity on that).
[Terretta]: While visiting not(USA) recently, found Xbox Cloud streaming no longer viable (lag fest) though last year travels (even over 5g hotspot) it was fine. Doesn't bode well for this streaming concept. Looked for a real console and discovered all small shops and big box chains offer Nintendo and PS, none have Xbox. Speaking with managers revealed they or their suppliers "lost" or "ended" contracts to carry it. Secondary market / used was dirt cheap – good condition c...
[ChrisRR]: Sounds fair. If you essentially already have a TV or other device as a thin client, I can see many people happily paying $12 for 50 hours compared to spending $1000 on the entire console
Symantec 披露“Skynet 模式”AI 智能体劫持无防护大语言模型发起自持续网络攻击
The Decoder：AI News（RSS） · 10-22 11:30 UTC
赛门铁克披露一种“Skynet 模式”AI 智能体，可劫持缺乏防护的大语言模型并发动可自我维持的网络攻击。
AI 判断：标题初筛（未依据正文）：标题仅披露安全研究结论，缺少普通人可复现的防护演示，偏技术安全议题。
展开原始摘录
Symantec 研究人员通过蜜罐截获一个把大语言模型嵌入核心的攻击智能体，它会劫持互联网上无防护的模型端点，自主扫描目标、利用服务器漏洞、窃取数据库凭据并安装挖矿程序，最多连续执行 100,000 步，算力费用由被劫持模型的主人承担。
openai/codex released rust-v0.164.0-alpha.12
github-actions[bot] · 10-22 11:11 UTC
据标题显示，OpenAI 的 Codex 项目在 GitHub 上发布了 Rust 版本 rust-v0.164.0-alpha.12 的 Alpha 预发布版本。
AI 判断：标题初筛（未依据正文）：仅凭标题可判断这是开发者命令行工具的常规 Alpha 版本更新，没有面向普通用户的新能力或可演示效果。
展开原始摘录
Release 0.164.0-alpha.12
Tell HN: I unknowingly broke the PHP RCE embargo. I am sorry.
petecooper · 10-22 10:11 UTC
一位 HN 用户发帖称自己无意中破坏了 PHP 远程代码执行漏洞的披露禁运规定，并为此道歉。
AI 判断：标题初筛（未依据正文）：该内容属于开发者安全披露话题，与普通人的 AI 使用和生活场景无关，且标题提供的信息有限。
展开原始摘录
I posted here about the then-upcoming PHP RCEs related to CVE-2026-109618 & CVE-2026-109619. I appreciate this has had considerable knock-on effects with the schedule for co-ordinated disclosure, and created additional workload.
I am truly sorry. I take full responsibility for my actions.
It was not remotely my intention to cause any of the subsequent side effects.
At the time of posting, I didn't know about any embargo. My contact had told me there was no embargo, and her contact had told her the same thing.
As someone who follows the 4-weekly PHP release cycle and compiles it from source, my intention was to alert others in a similar position to be aware of the importance/urgency of the updates.
That clearly backfired in a pretty wild fashion. I have learned my lesson.
--- Top Comments ---
[john_strinlai]: if you received the information about the RCEs without the information about the embargo, then you did not break the embargo . someone else up the chain did. this whole saga will be analyzed and we'll all come out better for it at the end.
[rancar2]: That’s kind of you, Pete, but it was leaked and being exploited well prior to your previous posting here: https://aws.amazon.com/blogs/security/amazon-threat-intellig...
[dabinat]: I don’t know if embargoes have much effect anymore. They used AI to find this issue, meaning anyone else could also find the issue with time and the right model. The clock is always ticking after a CVE but now it ticks faster. I am certain that companies like Cellebrite are pointing a bunch of models at popular repos and filing away all the exploits they find in a database.
[cwillu]: Don't be silly, you can't possibly be responsible for breaking an embargo you weren't subject to or informed about.
[qubex]: What does “take full responsibility” mean in the present context? Are you forfeiting all your assets and pledging the next twenty generations of your descendants into peonage? (For clarity, I’m being facetious, but it’s a really meaningless and empty phrase to commit to the written record.)
AI finds a security hole in 70% of the web
The Rundown AI · 10-22 10:00 UTC
标题称，AI 在 70% 的网页中发现了一个安全漏洞。
AI 判断：标题初筛（未依据正文）：标题仅有宽泛的安全统计，缺少普通人可操作细节，属于技术安全类。
展开原始摘录
PLUS: Let Town prepare your next meeting brief
Faster than earcut: triangulating 3.7M map polygons in 489 ms
molind · 10-22 09:31 UTC
一篇技术博客声称其地图多边形三角剖分方案在 489 毫秒内处理了 370 万个多边形，速度超过 earcut。
AI 判断：标题初筛（未依据正文）：这是纯几何计算与渲染性能的工程优化内容，与普通人的办公、学习或生活没有直接关系，也没有可演示的消费级效果，应按技术类低分处理。（仅依据标题判断）
展开原始摘录
--- Top Comments ---
[amelius]: Should mention the mother of all triangulators: https://www.cs.cmu.edu/~quake/triangle.html
[mourner]: Hey Evgen, Earcut author here — thanks for the writeup and a really cool library! I need to see if there are any tricks to borrow for Earcut's JS version. :) One thing to note is that I deliberately had to reject some of the potential optimizations to keep the library small, so that bundle size doesn't blow up — for most practical applications (e.g. Mapbox GL JS), it's already fast enough (and 4.1k min/gz). Curious how big the WASM build of Earstrip is. On the C++ side, Ma...
[morishuz]: awesome. the performance improvements looks impressive! i’m the author of Delaunay32, a constrained delaunay triangulator with polygon and hole support. it uses divide-and-conquer rather than incremental insertion, so im curious how it would compare with the delaunay baseline here. https://github.com/morishuz/delaunay32 i havent benchmarked on your corpus yet, and our current polygon api requires simple rings with non-touching holes. thanks for publishing the datasets, t...
[mistercow]: I've wondered for a long time why ear clipping + optimizations has won on triangulation over stuff like Seidel's trapezoidal method, which has O(nlog*n) time complexity and handles holes naturally. I know big-O isn't everything, but that's essentially linear versus quadratic, and it seems like we care about triangulating more and more complex polygons as time goes on. Is it just a triangle quality thing? Does Seidel lead to too many hard to deal with slivers?
AIBase 中文 · 10-22 09:19 UTC
标题称，一位海外AI创业者试用了千问AI硬件新品，并称其翻译效果令人难以置信。
AI 判断：标题初筛（未依据正文）：仅有标题信息，翻译硬件对普通人有潜在实用性，但缺少产品细节、价格与可验证演示，需谨慎判断。
展开原始摘录
AI资讯 AI新闻资讯 正文 “翻译效果令人难以置信”，海外AI创业者上手千问AI硬件新品 发布于AI新闻资讯 发布时间 : 2026年10月22号 17:19 阅读 : 1 分钟 千问AI眼镜N1系列及千问AI耳夹式耳机正式发售后，受到彭博社等美国媒体关注，其中有报道以“中国智能眼镜再次让Meta相形见绌”为题，基于公开规格对比双方的影像、眼动追踪等能力。 在旧金山Assembling2026活动期间，不少AI创业者及技术从业者也体验了这批新品。 “就像你在直接和我说话。”创业顾问Mike Rice戴上千问AI耳夹式耳机，工作人员说中文，他在耳边听到英文译文。自然的翻译体验，让他看到了与不同国家创业者顺畅交流的可能。 “它的音质让我很惊喜。”Envision联合创始人兼CEO Karthik Mahadevan表示，耳机的声音表现超出了他对开放式设计的预期。体验千问AI眼镜时，小型摄像头拍出的清晰照片也给他留下了深刻印象。 N1系列搭载5000万像素影像系统，N1Pro新增体征实时监测、眼动交互和虹膜核身。千问AI耳夹式耳机经过Bose专业调音，支持89种语言专业翻译，通过千问App中的“千问耳机”入口绑定后，还可语音交办任务、接收完成提醒和设置定时播报。 目前，三款新品已在淘宝、天猫、京东、抖音等渠道开售。双11期间，千问AI眼镜N1、G1配镜享7折，S1配镜享8折。 相关推荐 万象有声发布 Audimind，有声书制作工具出海 原懒人听书核心团队打造的万象有声推出海外版 Audimind（audimind.ai），定位有声书制作工具，支持演播者与团队在同一项目内完成文稿处理、真人录音或 AI 配音、对轨、后期、审听及团队分工。免费版每月赠 10,000 积分，个人版 19 美元/月起，团队版 59 美元/月起，产品已在 Product Hunt 上线。 2026年10月23号 15:30 179.8k Anthropic 上市前亮出特殊治理架构：把控制权交给不持股的外部受托人 Anthropic冲刺IPO，外界多关注其收入增速、模型能力与万亿估值预期，但更特殊的是治理安排：不同于硅谷常见的AB股双层股权，也不把权力交给投资人，而是将公司实际掌控权交给不持股、不分利润的外部受托人，前美联储主席伯南克即在其中。权力与股权、收益分离，成为其迥异于常规科技公司上市模式的关键点。 2026年10月22号 17:55 153.6k 王兴兴给宇树画了三条线：量产、小模型、大算力 2026世界物理AI大会暨宇树合作伙伴生态大会在杭州召开。宇树科技董事长王兴兴将公司未来打法拆为三条线：一是产能，强调提升机器人规模量产能力，瞄准千万级乃至亿级市场，从“能造几台”走向“能造几百万台”，以规模作护城河；二是走“自研+合作”的复合路径。 2026年10月22号 17:20 96.4k AI日报：讯飞星火X2.6发布；红果短剧上线小果AI；宇树发布H2X人形机器人测试广告 【AI日报】聚焦AI热点与开发趋势。科大讯飞发布星火X2.6端云一体大模型，基于国产算力训练，采用293B A30B MoE架构，上下文长度达1M，重点提升代码与智能体能力，推动智能体能力同步升级，并推出相关AI新功能。该模型展现国产大模型在算力、长上下文和智能体方向的新进展。 2026年10月22号 16:41 156.4k 讯飞星火 X2.6 端云一体大模型发布，首次基于全国产算力突破代码与智能体能力 科大讯飞在全球1024开发者节发布讯飞星火X2.6端云一体大模型，首次基于全国产算力，同时实现代码与智能体能力。董事长刘庆峰称，大模型公司中仅星火纯用全国产算力训练，其余几乎依赖英伟达，且讯飞模型尺寸最小。为提升国产算力效率，讯飞正持续优化。 2026年10月22号 16:15 155.5k 智启未来，您的人工智能解决方案智库 简体中文 © 2026 AIBase 备案号：闽ICP备08105208号-24
Jailbreaking the Switch 2
ledoge · 10-22 08:01 UTC
该文章标题称已破解 Switch 2，仅凭标题无法确认具体方法与影响。
AI 判断：标题初筛（未依据正文）：标题只涉及游戏机破解，与 AI 无关且偏技术，普通读者实用性低。
展开原始摘录
--- Top Comments ---
[hturan]: Incredible all around. If this is true decrypted `fs` r/w, I wouldn't be surprised if we start seeing Switch 2 game dumps appear in a few hours. Given the world of recomps/decomps we now live in, that would mean console exclusives playable on arbitrary hardware (a worst-case scenario for Nintendo). We've already seen things like Tears of the Kingdom running at 60fps natively on an iPhone. Giving LLMs arms and eyes via KVM (or webcam) is a very strong paradigm for this sort...
[altairprime]: Fair warning that Nintendo likes to remote-kill both Switch 2 consoles and the game cards used by them once they detect jailbreaks from all future accesses to online services, so you should probably take steps to permanently detach your Switch from all online services and from WiFi before experimenting with this, and expect that selling any game carts from your jailbroken console can land you in hot water with the buyer reporting your info to Nintendo someday. Thanks, secure attestation of ...
[valgaze]: Perfection: The JetKVMs allow Codex to send controller inputs to the Switches over USB and receive video output over HDMI. We also need a way to press the power button—when the Switch 2 crashes, it needs a button press to restart.
That’s what the Bluetooth light switch flicker from Amazon is for. When the lever comes down, it presses the power button. Once the power button is fully pressed, the rest of the lever’s torque lifts the mechanism.
[rootsudo]: I don't understand, how are people using codex for this sort of research? Whenever I try to do bug bounty, research, etc it locks down and I have the "blue" tier access/daylight whatever it's called. For example I tried to investigate some CVE's on older android roms and it downright was refusing it. It was okay when doing it on... Microsoft and to submit a microsoft bug bounty but android was no go.
[nonfunctional]: That contraption used to press the power button is the SwitchBot Bot[1], made by SwitchBot. That makes this one a "Switch SwitchBot Bot". https://www.switch-bot.com/products/switchbot-bot
openai/codex released rust-v0.163.0-alpha.11.2
github-actions[bot] · 10-22 06:41 UTC
openai/codex 发布了 Rust 版本 v0.163.0-alpha.11.2。
AI 判断：标题初筛（未依据正文）：从标题看，这只是 Codex 的开发者工具 Alpha 版本更新，对普通用户缺乏实用场景和可演示价值。
展开原始摘录
Release 0.163.0-alpha.11.2
Kling 4.0 上线 OpenArt，支持首中尾帧引导与最多 10 张参考图
X：可灵 Kling AI (@Kling_ai) · 10-22 06:32 UTC
标题显示，可灵 4.0 已在 OpenArt 平台上线，新增首帧、中间帧、尾帧引导能力，并支持最多 10 张参考图输入。
AI 判断：标题初筛（未依据正文）：这是视频生成工具一次具体的可控性升级，画面感强、便于演示，但标题未提及价格、地区开放与使用限制，实际可用性仍需核实（基于标题判断）。
展开原始摘录
OpenArt 宣布 Kling 4.0 上线，可用首、中、尾帧引导视频，最多使用 10 张参考图，并支持文本或帧生成与视频编辑，All-in-One Reference 用一个模型完成生成和编辑。11 月 20 日前 Pro 和 Wonder 套餐享 50% 折扣。
Swyx: ratioed @cognition lololololol never change twitter
Swyx · 10-22 06:09 UTC
Swyx 发推调侃 Cognition 的帖子在推特上被“ratio”（回复量远超点赞），并感叹推特风格依旧。
AI 判断：标题初筛（未依据正文）：仅凭标题可知这是一条网络调侃式推文，没有 AI 产品、用法或可核实信息，对普通观众无实用价值。
展开原始摘录
ratioed @cognition lololololol never change twitter
Zara Zhang: Your personal website is the new resume
Zara Zhang · 10-22 06:02 UTC
Zara Zhang 提出个人网站正在取代传统简历，成为新的个人展示方式。
AI 判断：标题初筛（未依据正文）：标题仅为个人观点，未涉及具体 AI 功能、教程或可验证案例，实用性有限。
展开原始摘录
Your personal website is the new resume
What FDE means at OpenAI
OpenAI News · 10-22 06:00 UTC
据标题看，OpenAI 发文解释其「前置部署工程」（FDE）岗位/模式的含义。
AI 判断：标题初筛（未依据正文）：仅凭标题判断：这是企业服务与岗位模式介绍，属工程组织话题，普通用户难有可演示或用得上的内容。
展开原始摘录
What forward deployed engineering means at OpenAI
爱范儿 · 10-22 05:58 UTC
标题称 PHP 出现严重漏洞，导致全球紧急「拔网线」式处置，并提醒不要指望 AI 来解决互联网安全问题（仅基于标题，细节未知）。
AI 判断：标题初筛（未依据正文）：这是面向开发者的底层组件安全事件，普通人难直接使用或复现，标题也未给出可演示的细节，按规则归为技术类并压低评分。
展开原始摘录
网站被攻破后，数据、密码和令牌都可能受到影响
#欢迎关注爱范儿官方微信公众号：爱范儿（微信号：ifanr），更多精彩内容第一时间为您奉上。
Peter Yang: If you enjoyed this, sign up for free to my newsletter to get my best AI and product guides....
Peter Yang · 10-22 05:51 UTC
这是一条引导读者免费订阅作者邮件通讯的推广贴，标题未说明具体 AI 内容或功能。
AI 判断：标题初筛（未依据正文）：仅标题可见，本质是邮件通讯订阅推广，没有可核实的新功能、教程或事件信息，不适合做成面向普通人的选题。
展开原始摘录
If you enjoyed this, sign up for free to my newsletter to get my best AI and product guides.
Join 110,000+ subscribers here:
https://t.co/yz9xcsfrhF
Garry Tan: Gabriel Zucman is an economic adviser to the union sponsoring California’s Prop. 40, a contri...
Garry Tan · 10-22 05:45 UTC
Garry Tan 称 Gabriel Zucman 是资助加州第40号提案的工会的经济顾问。
AI 判断：标题初筛（未依据正文）：该内容为美国政治与经济议题，与 AI 无关，不适合作为面向普通人的 AI 选题。
展开原始摘录
Gabriel Zucman is an economic adviser to the union sponsoring California’s Prop. 40, a contributor to the research underlying its revenue estimates, and a public campaigner for its passage.
His ideas will destroy California.
Garry Tan: Gabriel Zucman is currently employed by the University of California at Berkeley, a public un...
Garry Tan · 10-22 05:42 UTC
加里·谭在推文中指出，加布里埃尔·祖克曼目前受雇于公立大学加州大学伯克利分校。
AI 判断：标题初筛（未依据正文）：仅从标题看，这是关于学者任职机构的时政评论，与 AI 新功能、工具或可演示案例无关，不适合面向普通人的 AI 选题。
展开原始摘录
Gabriel Zucman is currently employed by the University of California at Berkeley, a public university that will be hurt severely due to shortfall of tax revenue to the state: to the tune of $39 billion net loss in aggregate
Perhaps when the tax doom spiral hits his department some people will finally say damn we shouldn’t have hired this guy
Kling 4.0 现已上线 OnSolo
X：可灵 Kling AI (@Kling_ai) · 10-22 05:36 UTC
可灵 Kling 4.0 已在 OnSolo 平台上线。
AI 判断：标题初筛（未依据正文）：标题仅说明可灵 4.0 上线 OnSolo，缺少功能、价格和实际效果信息，暂只适合做简讯式核验选题。
展开原始摘录
可灵（Kling）宣布 Kling 4.0 已在 OnSolo 上线。该模型支持单次生成 30 秒原生视频，快速动作和大运镜下保持稳定，立体声配更准的口型同步，并提供 21:9 宽屏，单个任务可叠加最多 15 个参考图和 10 个关键帧以保持角色、镜头与剧情一致。
Peter Yang: Just finished episode 1 of Edgerunners 2. It's actually kind of...happy?
Peter Yang · 10-22 05:33 UTC
Peter Yang 表示自己刚看完《边缘行者 2》第一集，并觉得这一集竟然有点……开心。
AI 判断：标题初筛（未依据正文）：仅凭标题判断，这是与 AI 无关的影视娱乐内容，不适合面向普通人的 AI 选题。
展开原始摘录
Just finished episode 1 of Edgerunners 2.
It's actually kind of...happy?
Aaron Levie: If you wanted to know all the jobs that AI will produce, look no further than cybersecurity....
Aaron Levie · 10-22 05:15 UTC
Aaron Levie 认为，AI 将创造的岗位可以从网络安全行业中得到充分体现。
AI 判断：标题初筛（未依据正文）：仅凭标题看，这是一条缺乏数据或可演示案例的观点引子，实用性和可核实性有限。
展开原始摘录
If you wanted to know all the jobs that AI will produce, look no further than cybersecurity. This is going to become one of the biggest investment areas for enterprises going forward.
And even as the work becomes more agentic and autonomous you will need more talent to point the agents at the most critical work, review findings, patch systems, respond to incidents, and more.
Lots of opportunity right now for the systems that protect data and enterprises, as well as the professionals that are deploying them and protecting organizations.
It’s going to only get more wild out there.
NSA 就 AI 发现的 PHP2Shell 关键漏洞发布修复警告
X：Rohan Paul (@rohanpaul_ai) · 10-22 05:13 UTC
美国国家安全局针对 AI 发现的 PHP2Shell 关键漏洞发出修复警告，提示相关系统尽快修补。
AI 判断：标题初筛（未依据正文）：仅凭标题可见这是面向网站运维与开发者的安全漏洞通告，普通用户难以直接使用或演示，属技术类选题。
展开原始摘录
NSA 警告两个由 Calif 研究人员用 GPT-5.6 Sol 发现的 PHP 关键漏洞（PHP2Shell），敦促组织打补丁或将未修补服务器下线。两个漏洞位于 PHP 核心，在 Apache mod_php 上演示的利用可在应用代码运行前触发。
anthropics/claude-code released v2.1.308
ashwin-ant · 10-22 04:34 UTC
据标题信息，GitHub 上 anthropics/claude-code 仓库发布了 v2.1.308 版本更新。
AI 判断：标题初筛（未依据正文）：仅为开发者命令行工具的常规版本发布，标题未提供任何普通用户可用的功能或场景。
展开原始摘录
## What's changed
- Bug fixes and reliability improvements
openai/codex released rust-v0.164.0-alpha.11
github-actions[bot] · 10-22 04:22 UTC
OpenAI 的 Codex 项目在 GitHub 上发布了 rust-v0.164.0-alpha.11 版本。
AI 判断：标题初筛（未依据正文）：这是面向开发者的命令行工具内部 alpha 版本更新，标题未体现任何普通人可用的功能或影响。（依据标题判断）
展开原始摘录
Release 0.164.0-alpha.11
Peter Yang: Securing all my accounts with AI's help tonight before something bad happens. Also finally st...
Peter Yang · 10-22 04:15 UTC
作者在推文中表示，自己今晚要借助 AI 保护所有账户，以防出现安全问题，并提到还终于做了另一件事。
AI 判断：标题初筛（未依据正文）：仅凭标题可知这是一条被截断的个人推文，既没有说明用了什么 AI 或什么方法，也没有可复现的步骤和效果对比，因此实用价值和信息密度都偏低。
展开原始摘录
Securing all my accounts with AI's help tonight before something bad happens. Also finally stopped being cheap and paid for @1Password
WSJ：马斯克希望以 AI 为核心推动 SpaceX 与 Tesla 合并
X：Rohan Paul (@rohanpaul_ai) · 10-22 03:16 UTC
据《华尔街日报》报道，马斯克有意以 AI 为核心推动 SpaceX 与 Tesla 合并。
AI 判断：标题初筛（未依据正文）：这是仅凭标题判断的商业合并传闻，未证实，也不含普通人可用功能或可演示内容，实用价值低。
展开原始摘录
WSJ 报道，马斯克希望以 AI 为核心将估值约 $2T 的 SpaceX 与约 $1.5T 的 Tesla 合并。SpaceX 的 AI 业务已成其最大业务，Q3 营收 $7.56B（一年前仅 $813M），季度 $19.6B 资本开支中超过 $16.4B 投向 AI，推动 SpaceX 实现季度利润 $4.4B（上年同期亏损 $1.05B）。
X：Rohan Paul (@rohanpaul_ai) · 10-22 03:13 UTC
据 Bloomberg 报道，DeepSeek 新一轮融资规模约为 150 亿美元。
AI 判断：标题初筛（未依据正文）：这是纯融资新闻，标题未提供普通人可用功能或可核实细节，实用性和演示性低。
展开原始摘录
据 Bloomberg，DeepSeek 最新一轮融资约 150 亿美元，为中国最大规模私募融资之一。谈判期间融资目标至少两次上调，从约 75 亿美元提高到 120 亿美元；CATL 和腾讯是最大投资方之一，CATL 据报出资约 15 亿美元。投前估值约 748 亿美元，加上 6 月完成的首轮超 75 亿美元，DeepSeek 约 6 个月内共募资约 225 亿美元。
Swyx: this is my quant https://t.co/9KGoY9XA17 https://t.co/d71VZ8oPLf
Swyx · 10-22 02:40 UTC
Swyx 发文称“这是我的 quant”，并附有两个链接，但标题未说明具体用途或功能。
AI 判断：标题初筛（未依据正文）：标题只有一句含糊的“我的 quant”，缺少可核实的功能、演示或普通人可用的信息，暂不适合作为教程或趋势选题。
展开原始摘录
this is my quant https://t.co/9KGoY9XA17 https://t.co/d71VZ8oPLf
AI 挖出潜伏十几年的 PHP 底层漏洞，PHP2Shell 事件全景复盘
X：卡兹克 (@Khazix0918) · 10-22 02:33 UTC
一则复盘称，AI 发现了潜伏十几年的 PHP 底层漏洞，该事件被称为 PHP2Shell。
AI 判断：标题初筛（未依据正文）：仅凭标题可见这是 PHP 底层安全漏洞话题，属于开发者与安全圈议题，普通人难以直接使用或演示（判断基于标题）。
展开原始摘录
作者复盘 PHP2Shell 事件：两个 CVSS 满分定为 9.2 分的 PHP 远程代码执行漏洞（CVE-2026-109618 和 CVE-2026-109619）在补丁发布前泄露，攻击演示视频外流，多国政府网站预防性下线，OpenAI 认领漏洞由 GPT-5.6 Sol 辅助发现。
IT之家·人工智能 · 10-22 02:17 UTC
有消息称，Anthropic 最快将于本周五公开提交 S-1 招股书，可能成为史上规模最大的 IPO。
AI 判断：标题初筛（未依据正文）：该消息属于 Anthropic 上市融资传闻，对普通人的工作与生活缺少可直接演示或使用的价值，且尚未证实。
展开原始摘录
据雅虎财经援引知情人士报道，Anthropic 最早或于本周五公开提交 S-1 招股说明书，推进有望创史上最大规模的 IPO。
Nikunj Kothari: Exhibit 347 of why you should have kids 🥰 https://t.co/bhSmo5jHD0
Nikunj Kothari · 10-22 01:41 UTC
这是一条标题仅为个人生活化调侃的推文，未提供可核实的 AI 新功能或事件信息。
AI 判断：标题初筛（未依据正文）：标题仅为个人生活化调侃，未提供可核实的 AI 新功能或事件信息，不适合作为 AI 选题。
展开原始摘录
Exhibit 347 of why you should have kids 🥰 https://t.co/bhSmo5jHD0
爱范儿 · 10-22 00:21 UTC
这是一份科技早报，汇总了iPhone Duo每人限购两台、雷军称汽车没有安全其他都是0，以及宝马员工争相离职挤崩系统等消息。
AI 判断：标题初筛（未依据正文）：仅凭标题看，这是多条科技新闻的汇编，没有明确的AI新功能或可演示内容，对普通人的实用价值有限（基于标题判断）。
展开原始摘录
· PHP 紧急修复两个严重漏洞，攻击者无需登录即可远程执行代码
· 三星、SK 海力士升级中国工厂，拟转产更高层数 NAND
· Today AI 累计融资超 2 亿美元，腾讯、蚂蚁等参投
#欢迎关注爱范儿官方微信公众号：爱范儿（微信号：ifanr），更多精彩内容第一时间为您奉上。
Zara Zhang: Made with just one prompt! The prompt: https://t.co/tDgYo8dVyB
Zara Zhang · 10-22 00:20 UTC
Zara Zhang 展示了一件只用一个提示词生成的作品，并附上了提示词链接。
AI 判断：标题初筛（未依据正文）：标题只说明单提示词生成作品并给出链接，缺少可核实的内容细节，更偏向展示型选题，而非完整教程。
展开原始摘录
Made with just one prompt! The prompt: https://t.co/tDgYo8dVyB
公众号：数字生命卡兹克 · 10-22 00:06 UTC
文章介绍卡兹克对 AI 发现的 PHP2Shell 漏洞的详细解读，指出发现仅用两天而修复耗时两个月。
AI 判断：标题初筛（未依据正文）：标题显示这是偏技术安全漏洞话题，对普通人的直接实用性和可演示性较弱，仅基于标题判断。
展开原始摘录
作者复盘 PHP 两个 CVSS 满分起步的远程代码执行漏洞（CVE-2026-109618、CVE-2026-109619，合称 PHP2Shell）从提前泄密、演示视频流出、多国政府网站预防性下线，到21日晚10点补丁上线的全过程，并指出这两个漏洞是 GPT-5.6 Sol 辅助发现的，OpenAI 还发布了检测 skill。
openai/codex released rust-v0.164.0-alpha.10
github-actions[bot] · 10-22 00:04 UTC
OpenAI 的 Codex 项目发布了 Rust 版本 v0.164.0-alpha.10。
AI 判断：标题初筛（未依据正文）：标题显示这是面向开发者的预发布版本更新，普通用户缺少可直接使用或演示的场景。
展开原始摘录
Release 0.164.0-alpha.10
Uber brings ChatGPT Work to all employees with OpenAI
OpenAI News · 10-22 00:00 UTC
Uber 与 OpenAI 合作，将 ChatGPT Work 推广至公司全体员工。
AI 判断：标题初筛（未依据正文）：仅凭标题判断，这是企业内部的 AI 工具部署，普通用户无法直接使用，实用性和可演示性有限。
展开原始摘录
Uber is rolling out ChatGPT Work to more than 30,000 employees, helping teams research, analyze data, create presentations, and complete complex workflows.
Anthropic 指南：何时把子智能体合并进主 Agent 循环
Claude：Blog（网页） · 10-22 00:00 UTC
Anthropic 发布指南，讨论构建智能体时何时应把子智能体合并进主 Agent 循环。
AI 判断：标题初筛（未依据正文）：这是面向开发者的智能体架构指南，缺少普通人可直接使用的功能或可演示效果，按规则归为技术类。
展开原始摘录
Anthropic Applied AI 团队发布指南，讲解何时把 orchestrator 下的路由器、规划器和领域子智能体合并进单个主 Agent 循环。文中对比显示，带 skills 的单智能体在质量上常优于一提示全包和子智能体设计，且成本和延迟更低，一家客户改用单循环后运行时间约快 2x。
Nikunj Kothari: Who knew acquiring super intelligence could lead to the entire fabric and trust of the intern...
Nikunj Kothari · 10-21 23:57 UTC
Nikunj Kothari 发推感叹，获取超级智能可能影响整个互联网的信任基础。
AI 判断：标题初筛（未依据正文）：标题仅为模糊的感想式推文，没有事实、产品信息或可演示内容，不适合作为普通用户的实用选题。
展开原始摘录
Who knew acquiring super intelligence could lead to the entire fabric and trust of the internet being slowly broken..
It’s truly wild and batsh!t bananas out there. https://t.co/gA5AQr2s6b
Zara Zhang: Turn a photo into a 3D popup card https://t.co/t4YdSAPIIB
Zara Zhang · 10-21 23:51 UTC
Zara Zhang 分享了一项把照片制作成 3D 立体贺卡的演示。
AI 判断：标题初筛（未依据正文）：标题显示可把普通照片变成有画面感的 3D 立体贺卡，适合做创意效果演示，但具体工具和步骤尚不明确。
展开原始摘录
Turn a photo into a 3D popup card https://t.co/t4YdSAPIIB
Guillermo Rauch: Vercel has proactively rolled out Firewall-level mitigations for today's Node.js RCE vulnerab...
Guillermo Rauch · 10-21 23:29 UTC
Guillermo Rauch 表示，Vercel 已针对当日披露的 Node.js 远程代码执行漏洞，主动上线防火墙层面的缓解措施。
AI 判断：标题初筛（未依据正文）：这是面向开发者的基础设施安全动态，普通用户难以感知或使用；仅凭标题也无法核实漏洞细节与影响范围。
展开原始摘录
Vercel has proactively rolled out Firewall-level mitigations for today's Node.js RCE vulnerability, in advance of the official patch release.
All traffic for all deployments is automatically protected. We're actively monitoring and rolling out managed runtime upgrades. https://t.co/fSgaLf99o7
