爱范儿 · 10-27 22:37 UTC
谷歌发布 Genie 4，标题称 120 秒的世界让人看见 AI 的未来。
AI 判断：标题初筛（未依据正文）：标题仅显示 Genie 4 发布与未来展望，普通人的直接用途不明确，判断基于标题。
展开原始摘录
AI 的下一个目标是模拟现实
#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。
从 62 条内容中筛选出 6 条重要资讯。
- AI 日报：即梦 AI 上线短剧工作室；蚂蚁灵光 2.0 发布；千问 App 上线学习功能 ⭐️ 7.1/10
- 苹果 iOS 27.2 接入谷歌 SynthID，为 AI 图片添加隐形水印 ⭐️ 7.0/10
- OpenAI 发布 GPT-6.1 Terra，标准 API 价格较上代降低 50% ⭐️ 6.85/10
- DeepMind 发布 Genie 4：实时交互世界模型，原生生成音频 ⭐️ 4.5/10
- 微软发布 MAI Image 2.7 图像编辑模型，称 Artificial Analysis 排名第二 ⭐️ 4.4/10
- 月之暗面发布开源旗舰模型 K3.1 Pro：2.8T 参数、支持百万上下文 ⭐️ 2.9/10
AI 日报：即梦 AI 上线短剧工作室；蚂蚁灵光 2.0 发布；千问 App 上线学习功能 ⭐️ 7.1/10
AI 日报 2026 年 10 月 27 日汇总：即梦 AI 上线“短剧工作室”，称整合剧本整理、分镜生成、视频制作与后期，DreamScript 支持 10 万字文本和 100 集生成，实际质量与长剧本连贯性待验证；蚂蚁灵光 2.0 发布，主打跨会话记忆、长程任务调度并接入支付宝生态，称上线 6 天下载破 200 万，隐私授权与稳定性待验证；千问 App 上线找题、拍题答疑及学习档案功能，提分效果需长期验证。
web · AIbase AI 日报 · 10月27日 07:51
苹果 iOS 27.2 接入谷歌 SynthID，为 AI 图片添加隐形水印 ⭐️ 7.0/10
据 MacRumors 报道，苹果已在 iOS 27.2 中接入谷歌的 SynthID 技术，为“图乐园”生成的图片及经 Apple 智能编辑的部分照片添加肉眼不可见的像素级水印。谷歌称该水印不损伤画质，并能承受裁剪、滤镜和压缩。资料称，使用“照片”App 的“扩展”和“重新构图”编辑后，检测时会判为 AI 生成；“清理”功能是否添加水印取决于改动幅度，而“图乐园”产出图片均自带 SynthID。iPhone 暂不提供检测入口，用户需通过谷歌 SynthID Detector 网页验证。上述效果来自报道所述测试，苹果与谷歌的正式说明及更广泛兼容性尚未明确。
web · AIBase 中文 · 10月27日 08:29
OpenAI 发布 GPT-6.1 Terra，标准 API 价格较上代降低 50% ⭐️ 6.85/10
据爱范儿报道，OpenAI 发布新一代模型 GPT-6.1 Terra，主打更快响应、视觉理解与电脑操作能力，定位为低成本 Agent 工作模型。OpenAI 称，其标准 API 价格较 GPT-5.6 Terra 降低 50%，输入每百万 Token 1 美元、输出 5 美元；在 Agents’ Last Exam 等测试中表现提升，事实错误率下降，编程成绩翻倍；部分速度与成本数据尚未验证。
rss · 爱范儿 · 10月27日 05:00
DeepMind 发布 Genie 4：实时交互世界模型，原生生成音频 ⭐️ 4.5/10
Google DeepMind 于 10 月 27 日发布世界模型 Genie 4，称其原生多模态，可在生成实时交互环境时同步产出音乐、环境声和人物对话，并支持用自然语言实时改变模拟。模型基于 Gemini Omni 等 Gemini 模型维持场景一致性与真实感。官方称可用于机器人训练等场景；目前仅向可信测试者开放研究预览，后续计划通过 Project Genie 提供给 Google AI Ultra 订阅者，并向 Gemini API 和 AI Studio 早期合作伙伴开放。部分评论指出，物体离开画面再返回时细节会变化，音频示例中存在乐器误判。
hackernews · Future Tools · 10月27日 16:06 · 社区讨论 · 中文阅读 · 4 个来源
微软发布 MAI Image 2.7 图像编辑模型，称 Artificial Analysis 排名第二 ⭐️ 4.4/10
微软发布图像编辑模型 MAI Image 2.7，目前以预览形式推出。微软称其在 Artificial Analysis 图像编辑榜排名第二、Arena Edit 排名第三，并使微软 AI 成为排名第二的实验室。该模型在 Artificial Analysis 评估的全部 10 类用例和 7 项编辑操作上均优于 2.6 版，在增强与修复、基于推理的编辑、场景与风格、文字或符号编辑达到类别前沿，其中场景与风格领先次优模型 24 Elo。上述排名和对比均为微软公布信息。
web · Future Tools · 10月27日 16:00
月之暗面发布开源旗舰模型 K3.1 Pro：2.8T 参数、支持百万上下文 ⭐️ 2.9/10
据 AI 资讯报道，月之暗面发布开源旗舰大模型 K3.1 Pro，厂商称其参数规模达 2.8T，支持百万 Token 上下文。官方确认完整模型权重预计 11 月 4 日前释出，模型已上线 OpenRouter。报道称该模型在 Artificial Analysis Intelligence Index 上获 55 分、位列第四。标准版 API 输入每百万 Token 13 元、输出 68 元，称较 K3 成本近半；Highspeed 版输出速度为标准版 2 至 3 倍、价格翻倍。上述性能与价格信息来自厂商及报道，尚待独立核验。
web · AIBase 中文 · 10月27日 08:28
更多资讯，供你自行挑选
以下内容已成功抓取，因评分、数量限制或同题合并未进入精选。AI 判断仅供参考，你可以继续阅读并自行挑选；各来源切换是全部列表的子集。
全部未入选
X：Anthropic (@AnthropicAI) · 10-27 22:20 UTC
Anthropic 宣布通过 OneGov 协议，以六折价格为美国各级政府提供 Claude。
AI 判断：标题初筛（未依据正文）：这是美国政府采购折扣消息，与普通中国用户的日常使用关联较弱，且目前仅有标题可核实。
展开原始摘录
Anthropic 宣布扩大与 USGSA 的合作，根据新的 OneGov 协议，以 40% 折扣向联邦、州、地方、部落及领地各级政府提供 Claude，详情见官方公告。
anthropics/claude-code released v2.1.312
ashwin-ant · 10-27 20:24 UTC
Claude Code 发布 v2.1.312 版本更新，主要包含网关身份识别配置、全屏队列消息点击发送，以及 VS Code 登录后策略拉取和大型 CLAUDE.md 文件被覆盖等问题的修复。
AI 判断：这是开发者命令行工具的版本更新日志，涉及网关配置和文件读写修复，普通用户难以直接使用或感知，属于纯技术内容。
展开原始摘录
## What's changed
- Added `oidc.user_id_claim` to the Claude apps gateway, to identify people by an id_token claim other than `sub`, such as Microsoft Entra ID's `oid`
- Added click to send: in fullscreen, clicking the "to send now" hint under a queued message sends it right away, as its key does
- Fixed managed settings and organization policy not being fetched for the rest of the session after signing in from the VS Code extension
- Fixed Claude overwriting a very large CLAUDE.md, rules file or @-included file in a subdirectory after seeing only its start; Write now asks Claude to read it all or edit in place
- Fixed per-directory and `paths:` skills not being found after a Bash `cd` deeper into a git repository
- Fixed a rare case where the first reply after resuming a session in the terminal could be missing the next time the session was resumed
- Fixed a command with an invalid name stopping all commands from loading, which could leave `claude -p` with no output
- Fixed the Read tool sometimes failing on PDF pages with "pdftoppm failed: Could not write image" when the clean-up of old session files ran at the same time
- Fixed the spinner, OpenTelemetry `effort` values and transcripts naming a newly picked effort level during a task before the model applied it; they now show the level in force
- Fixed the `/model` and `/effort` cache warning saying the re-read waits for your next message; a change made while Claude is working can trigger it sooner
- Fixed some sessions showing as waiting on you when the last message only mentioned items waiting on you mid-sentence
- Fixed agent view restoring input text on the next launch after it was cleared with Ctrl+C just before exiting
- Fixed agent view reopening in shell mode after exiting with Ctrl+D from an empty `!` input
- Fixed vim mode `d`, `c` and `y` with a `gj` or `gk` count larger than the lines left deleting or copying text; like Vim, they now only move the cursor
- Fixed `/verify` showing a long description meant for Claude in `/help` and the `/` command menu instead of a short one
- Improved the prompt's effort indicator: a change made mid-task now says it may wait for your next message, and the note clears when it no longer applies
- Improved WebFetch errors in cloud sessions: a host that does not resolve is now explained as possibly outside the allowed network domains, so Claude names the host to allow
- Changed Bedrock, Vertex, Foundry and Mantle to use the shorter system prompt for model names Claude Code doesn't recognize, such as custom deployment names and inference profile ARNs, even when they front Sonnet or Haiku; set `CLAUDE_CODE_SIMPLE_SYSTEM_PROMPT=0` to keep the full prompt
- Changed automatic model switches after a flagged message to also cover model names Claude Code doesn't recognize, such as custom deployment names; set `CLAUDE_CODE_DISABLE_REFUSAL_FALLBACK=1` to turn them off
- Changed usage limit messages to call the 5-hour window the "5-hour limit" instead of the "session limit"
- Changed cloud sessions started from the CLI to get their names and branch names from the server
- Changed prompt and agent hooks to run at most 24 at a time per session, sub-agents included, plus at most 12 that have waited longer than their event's timeout; a `timeout` under 1 second now counts as 1 second
- Changed a subagent's `autoCompactWindow` to lower the window set by `CLAUDE_CODE_AUTO_COMPACT_WINDOW`, where it had no effect before; it still never raises that window
- Changed OpenTelemetry `mcp_server.name` to report the real name of fixed-URL remote MCP servers declared by plugins from the official Anthropic marketplace, instead of `custom`
- Changed Claude Code to stop sending the `claude_code.safety_event` OpenTelemetry event
- [VSCode] Fixed sessions ending on a bare exit code when your organization's settings couldn't be loaded or were refused at startup; the tab now says why and offers Retry or Sign in again
- [VSCode] Changed usage limit resets to show a clock time ("Resets at 3:10 PM") and to say "5-hour limit"
- [Cloud sessions] Added an effort level to the model picker when you create or edit a routine that runs in the cloud
- [Cloud sessions] Added a Bash editor with colors, line numbers and a Copy button to the setup script box in cloud environment settings
- [Cloud sessions] Improved the step shown when Claude adds a repository to a cloud session: opening it now shows a short table of what happened instead of raw JSON
- [Cloud sessions] Fixed resending an unchanged message after a failed send in a cloud session sometimes running it twice
- [Cloud sessions] Fixed a failed cloud environment creation showing its error only as a toast outside the open dialog; it now appears inside the dialog
- [Cloud sessions] Fixed Edit in a routine's menu on the Routines page going to the routine's page without opening its edit dialog
- [Remote Control] Fixed an unarchived Remote Control session showing as working until a computer reconnected; it now shows as idle
- [Claude Tag] Improved Claude Tag's retry notice in Slack to name the model in use, or say rate limits were hit, instead of a generic API error
- [Claude Tag] Fixed a Claude Tag channel's on/off switch and agent settings falling back to workspace defaults when an older private Slack channel goes public and gets a new ID
- [Claude Tag] Fixed Claude in Slack not replying in channels whose environment has network access set to None
- [Claude Tag] Fixed Claude in Slack sometimes posting a file in the wrong thread when it was answering in another thread of the same channel
- [Code Review] Fixed Code Review reviewing an outdated commit, or none at all, when GitHub delivered a push event late; it now reviews the pull request's current commit
openai/codex released rust-v0.165.0
github-actions[bot] · 10-27 20:09 UTC
OpenAI 的 Codex 命令行工具发布 rust-v0.165.0，新增 Google Cloud Vertex AI 提供方、模型目录条目和会话导航等开发者功能。
AI 判断：这是面向开发者的命令行工具版本更新，普通用户难以直接使用，也缺少可演示的消费级场景。
展开原始摘录
## New Features
- Added a built-in Google Cloud Vertex AI provider, configured with `model_provider = "gcp"`. (#56265)
- Added GPT-6.1 Terra to the model catalogs. (#57337)
- Added a repository filter to the session picker and directory grouping to the agent command center, making it easier to navigate sessions across linked worktrees. (#56308)
- Added experimental attached thread forks to the app-server API. (#56351)
- Enabled reasoning effort changes during a conversation by default for supported models. (#56163)
- Enabled memory v2 by default for fresh profiles and added support for importing external-agent memories into it. (#56211, #56414)
## Bug Fixes
- Fixed fresh TUI sessions to honor the server’s default approval reviewer when no explicit override is provided. (#56090)
- Kept TUI preferences editable and immediately applicable when execution settings are invalid. (#56091)
- Cancelled remote MCP tool requests when their local response futures are dropped. (#56132)
- Prevented queued agent messages from restarting work after an interrupt or a switch to Plan mode. (#56330)
- Stopped retrying when a provider reports that a model was not found, preserving its error message. (#56445)
- Closed a marketplace URL validation gap that could allow Git to connect to a different host than the one checked against allowed patterns. (#56263)
## Changelog
Full Changelog: https://github.com/openai/codex/compare/rust-v0.164.0...rust-v0.165.0
- #56081 Cache decrypted Codex auth and managed secrets @celia-oai
- #56090 Honor server approval reviewer defaults on fresh TUI startup @etraut-openai
- #56091 Keep TUI preferences editable when execution settings are invalid @etraut-openai
- #56097 Prepare TUI startup to accept multiple backend sessions @bc-openai
- #56124 Forward Guardian reviewer events to parent raw-event subscribers @jif-oai
- #56127 Avoid blocking EMA auth status checks on account refresh @jif-oai
- #56132 Cancel MCP tool requests when response futures are dropped @rhan-oai
- #56133 Defer agent mail after empty final answers @eknight-oai
- #56139 Fix the MCP auth status test's IdP OAuth configuration @jif-oai
- #56153 Preserve safe enterprise MCP authentication error details @jif-oai
- #56154 Add managed network review domains before allow rules @olliem-oai
- #56163 Enable reasoning effort overrides by default @felixxia-oai
- #56182 Add AWS SigV4 signing to network proxy MITM hooks @viyatb-oai
- #56210 Stabilize the marketplace staging cleanup test @felixxia-oai
- #56211 Default fresh memory profiles to v2 @jif-oai
- #56214 Remove the `u64` cast from WSL DrvFS detection @felixxia-oai
- #56220 Use cargo_bin in the Windows sandbox manifest test @felixxia-oai
- #56222 Consolidate Bazel CI setup into `prepare-bazel-ci` @anp-oai
- #56224 Reuse current model context across paginated resume config loading @charliemarsh-oai
- #56233 Make Opus encoder test timestamps follow sample time @goodman-oai
- #56234 Emit prompt telemetry only after hooks accept input @abhinav-oai
- #56236 Make the code mode yield/resume test yield explicitly between phases @goodman-oai
- #56246 Bump Bazelisk to 1.30.0 in the Bazel CI setup action @anp-oai
- #56247 Consolidate Rust unit tests under dedicated test modules @dkovalenko-oai
- #56248 Bump uv to 0.13.5 in CI and release workflows @anp-oai
- #56250 Avoid redundant settings checkpoints when resuming threads @owenlin0
- #56261 Isolate V8 source dependencies and their C++ runtime settings @seanh-oai
- #56263 Reject backslashes in HTTP(S) marketplace URL userinfo @willwang-openai
- #56265 Add a built-in Google Cloud Vertex AI provider @celia-oai
- #56266 Preserve MCP disable reasons across runtime overlays @mzeng-openai
- #56272 Upgrade Python SDK model generation to datamodel-code-generator 0.64.0 @anp-oai
- #56274 Wait for cloud config refresh cancellation in loader lifetime tests @goodman-oai
- #56277 Support bounded request bodies in AWS SigV4 signing @viyatb-oai
- #56278 Add negotiated network policy metadata support for MCP calls @viyatb-oai
- #56280 Group consecutive transcript prompts into one background block @bc-openai
- #56284 Scope model catalogs and caches by thread originator @kennyyi-oai
- #56286 Bump `@modelcontextprotocol/sdk` to `1.32.0` @anp-oai
- #56288 Pin Linux artifact signing to Cosign v3.1.4 @anp-oai
- #56290 Lock Python release build tools with uv @anp-oai
- #56292 Fix MCP cancellation tests to use the metadata callback @anp-oai
- #56293 Use fixed-token auth fixtures in Vertex tests @anp-oai
- #56295 Decouple managed configuration loading from ChatGPT authentication @smurching-oai
- #56297 Mark built-in dummy credentials and strengthen collision checks @viyatb-oai
- #56301 Add TLS forwarding tests for MITM SigV4 requests @viyatb-oai
- #56302 Add hook configuration to environment attachments @andrewgu-oai
- #56304 Add support for environment-scoped managed hooks @andrewgu-oai
- #56308 Separate directory and repository scopes in the TUI @bc-openai
- #56309 Support model-specific writable-root permission wording @goodman-oai
- #56312 Track accepted conversation history in rollout traces @cassirer-openai
- #56313 Apply model catalog overrides to history, notes, and context tools @rhan-oai
- #56314 Add managed configuration loaders for custom config servers @smurching-oai
- #56330 Make agent-message wakes respect interrupts and Plan mode @alishobeiri-oai
- #56342 Match rollout trace image content across request snapshots @cassirer-openai
- #56343 Run environment hooks on their owning executors @andrewgu-oai
- #56351 Add experimental attached thread forks @btraut-openai
- #56364 Allow callers to set the thread source for sampling @keyz
- #56371 Make the Guardian review timeout configurable @vivi
- #56414 Support memory v2 in external-agent memory imports @jif-oai
- #56439 Update `ctutils` from 0.4.2 to 0.4.5 @felixxia-oai
- #56445 Treat model-not-found provider errors as terminal failures @charliemarsh-oai
- #56453 Make the Realtime V2 background-agent steering test deterministic @felixxia-oai
- #56454 Make sandbox mount cleanup tests tolerate file descriptor reuse @felixxia-oai
- #56459 Add Guardian V3 feature configuration @felixxia-oai
- #56460 Add Guardian v3 local board helpers @felixxia-oai
- #56461 Add silent append support to message boards @felixxia-oai
- #56462 Prevent memory imports from racing with consolidation @jif-oai
- #56465 Add experimental thread memory summary reads @jif-oai
- #57337 [0.165.0-alpha.1.2] Backport GPT-6.1 Terra to the model catalogs @shijie-oai
X：Gemini (@GeminiApp) · 10-27 20:00 UTC
Gemini 集成了 Google SynthID 检测功能，可用来判断图片、视频和音频是否由 Google AI 生成。
AI 判断：标题初筛（未依据正文）：据标题，这是面向普通用户的 AI 内容核验功能，实用但仅限 Google AI 生成内容，适合演示并说明局限。
展开原始摘录
Google Gemini 宣布集成 SynthID 检测功能，用户只需在 Gemini 中询问“Is this AI generated?”，即可检查图片、视频或音频文件是否由 Google AI 生成。官方称此举帮助用户更有信心地浏览网络内容。
Anthropic's Claude Accelerates Drug Discovery With 6 Biotech Partners, Launches Own Programs
Future Tools · 10-27 20:00 UTC
Anthropic 宣布其生命科学团队与 BigHat、百时美施贵宝等六家生物科技机构合作，用 Claude 加速药物研发，并称一个月内生成 3.5 万个食物过敏抗体设计、把 Enveda 的监管审查时间缩短 89%。
AI 判断：这是企业科研合作层面的成果，普通人既无法直接使用也难以核实，缺少可演示的功能与明确的个人收益。
展开原始摘录
Company News · 3h ago Anthropic's Claude Accelerates Drug Discovery With 6 Biotech Partners, Launches Own Programs Anthropic's life sciences team has partnered with six biotech organizations — BigHat Biosciences, Bristol Myers Squibb, Divergence Labs, Enveda, Generate Biomedicines, and The Column Group — to accelerate drug development using Claude. Key results include Claude generating 35,000 antibody designs for food allergy treatments in one month, reducing Enveda's regulatory review time by 89%, and helping improve cancer-targeting molecules to bind over 10 times more tightly. Anthropic also acquired Mithril Medicine and is running its own programs targeting rare muscular dystrophy, with clinical trials potentially starting by late 2028.
Tailscale SSH: Investigation and remediation update
figmert · 10-27 19:55 UTC
Tailscale 公布其 SSH 功能相关安全事件的调查结果与修复措施。
AI 判断：这是面向开发者和运维的网络安全事故通报，普通用户既难以理解也难以使用。
展开原始摘录
--- Top Comments ---
[minimaltom]: Former tailscalar here and now someone managing tailscale at an org. Having done my own review of the vulns and the fixes as they are landing in github, its pretty clear to me that this only affects situations where you use ACLs to permit ssh into a node (via tailscale ssh, not via opensshd), but dont give a path to sudoers/root. If a user had no SSH access and/or connectivity to a vulnerable node, none of these vulns give them access. However, if you have a vulnerable tailscaled an...
[rietta]: I am fairly glad I have stuck with stock OpenSSH that ships with Linux. This incident proves I am glad with that choice. Over the decades, I have maintained remote access through port forwarding on my router - at one point in 2008 I put SSH on port 80 so I could access it from the restricted wifi network at the hospital my grandmother was at. I have also used openvpn. I have also used cloudflare tunnels. But in the end, I always connect to hardened openssh with key only authentication.
[aborsy]: There are too many features, and lines of code (probably more than OpenVPN). I'm pretty sure there are all kinds of zero days in it, with this attack surface and feature set. I use it extensively, and I would rather pay them a fee if they could review thr code. A possibility is a lock down mode, disabling risky features. I just want a VPN for connrctiviy, no taildrop, taildrive, SSH, funnel, services, DNS, node sharing, ...They are disabled in my admin console, but the code runs on my de...
[siva7]: I wouldn't be too surprised if tailscale has become a critical vector of the internet like cloudflare over the years as most ai companies seem to use it
[wrxd]: I like Tailascale as a product but I decided to move my homelab to plain WireGuard. I wasn't using any of the additional features, I can get away with exposing a port on my router and I am changing devices rarely. News like this make me happy I removed a lot of complexity and attack surface
Survival LLM: AI assistant and emergency library that runs on your own computer
Luc · 10-27 19:55 UTC
一款可在个人电脑本地运行的生存主题 AI 助手，内置应急资料库。
AI 判断：本地离线应急助手概念新奇，但受众偏窄、部署门槛和实际效果仍待核实。
展开原始摘录
--- Top Comments ---
[x3haloed]: This topic is bringing out the hallucination-panic crowd. It's perfectly safe to use LLMs to surface underlying data and knowledge without having to trust that it's giving you correct advice. Use the LLM as search technology: string match -> semantic search -> search intelligence. This is especially helpful when you don't know the right question to ask. Semantics only help you find text that is semantically near the input query. Semantics won't help you when what you...
[jvanderbot]: Shoutout to project nomad. https://www.projectnomad.us/ its got a ton of useful options, preconfigured, integrates pretty easily, and can use any local LLM or data.
[fathermarz]: Emergencies = No Internet What about no power? You gonna run your generator for your local model? What are we doing here? Have we lost our way completely?
[rnd0]: Considering the tendency even now of LLMs to hallucinate, and the hardware requirements for any kind of LLM -why not just gather those resources into some sort of local archive that can be searched without an LLM? This seems redundant and requiring more overhead while providing less reliability than something like Kiwix to me.
[saaaaaam]: > Mac or PC with 16–31 GB | IBM Granite 4.2 8B | 14 of 22 right | 6 GB So on a regular consumer machine this is going to be wrong nearly half the time. Great! Let’s hope it’s only on the simple stuff and not the “how do I perform an emergency appendectomy”. Honestly, just buy some books.
Game Boy Color on My Conference Badge
Simon Willison · 10-27 19:34 UTC
Simon Willison 介绍在会议胸牌上运行 Game Boy Color 的趣味项目。
AI 判断：标题初筛（未依据正文）：这是硬件与模拟器方向的趣味技术项目，标题未体现普通人可用的 AI 功能或 AI 影响。
展开原始摘录
Game Boy Color on My Conference Badge
I'm at GitHub Universe this week and,
similar to last year
, the conference badge is a full-blown color screen computer.
I've been hacking around with it again, and then I ran into with Steve Francia... who worked with Gemini to get a
full NES emulator
working on it, and then went a step further and deployed an emulator for both the Game Boy and the Game Boy Color.
Which means, if you can obtain the ROM, you can run
The Legend of Zelda: Link's Awakening
on your conference badge!
A neat feature of the badge is that you can turn it upside down and then use game controller touch buttons that are built into the badge body.
Tags:
games
,
github
,
hardware-hacking
,
ai
,
generative-ai
,
llms
,
gemini
,
steve-francia
Discord will soon let anyone find public server messages through Google
DeepLogin · 10-27 18:44 UTC
Discord 宣布将让可发现公开服务器中的消息更容易被搜索引擎收录，用户可通过 Google 找到这些公开内容。
AI 判断：这是 Discord 的公开内容搜索可见性政策，并非 AI 新功能，主要影响 Discord 用户的内容隐私与曝光。
展开原始摘录
--- Top Comments ---
[SOTGO]: I'm tentatively hopeful. I think it's a disgrace how many communities have moved from the open internet into Discord, and at least this is continuing to move towards fixing a lot of the problems that migration caused. I do wonder why they chose now though, it seems like there's potentially a lot of value in keeping this data private, or at least selling it for a fee
[2001zhaozhao]: I'm a Discord server admin and I'd love to turn this on because having more discussion of my game indexable by search engines seems like a good thing in general. For new messages it seems like there's no problem, in combination with the opt-out option. But for existing users, how am I supposed to disclose the change to people? What if people posted content in the past and wouldn't want it on Google, but aren't active anymore? Furthermore, what if someone wants to opt-...
[CM30]: Not a fan of it requiring a certain number of members/level of activity/server status. My experience is that many of the communities hit hardest by Discord's login requirements aren't all that large, like speedrun communities for niche games, modding and ROM hacking communities for certain games, etc. I feel like it should be opt in/out regardless of the size of the server and amount of activity it has.
[freediver]: Just to note that Google is not mentioned in the title and indeed nowhere in the article. Searching the web is broader than 'google'.
[nixpulvis]: This is needed , but likely not enough yet. With the rise of Discord servers as community forums, we've lost the wealth of Q&A they used to provide. Being able to search for a user with the same issue as you and find it was no small part of what's let us all learn to program in the first place, as well as critical to the training data that powers the AIs we use to answer similar questions more quickly. There will always be new questions, one that even AI cannot figure out alone...
Claude AI Finds Critical RCE Exploits in Chrome, Safari, Android, Windows & Linux
Future Tools · 10-27 18:00 UTC
Anthropic 前沿红队使用未发布的内部 Claude 模型，在 Chrome、Safari、Android、Windows 和 Linux 中发现并利用多个严重漏洞，相关漏洞均已负责任披露。
AI 判断：这是使用未发布模型的网络安全研究，普通用户难以复现或直接受益，偏技术内容。
展开原始摘录
Research · 5h ago Claude AI Finds Critical RCE Exploits in Chrome, Safari, Android, Windows & Linux Anthropic's Frontier Red Team used an unreleased internal Claude model to discover and exploit critical vulnerabilities across Chrome, Safari, Android, Windows, and Linux. Findings include a one-click RCE chain in Chrome involving 62 patched CVEs, a Safari same-origin policy bypass, an Android lock screen bypass on a Pixel 9, a remote denial-of-service in Windows QUIC networking, and a zero-click Linux kernel RCE exploitable in under an hour. All vulnerabilities were responsibly disclosed and patched by the respective vendors.
Grok Bot Gets Pages, Vault, Apple Integrations, Video Creation, and More in Latest Update
Future Tools · 10-27 17:00 UTC
Grok Bot 最新更新加入 Pages、Vault、Apple 服务与 1Password 集成、视频创作等多项功能。
AI 判断：来源仅列出更新功能并提到演示视频，缺少具体用法、价格和地区限制，实用价值仍需核实。
展开原始摘录
Tools & Products · 6h ago Grok Bot Gets Pages, Vault, Apple Integrations, Video Creation, and More in Latest Update Grok Bot has received a significant update introducing several new features, including Pages, a Vault for storing information, and integrations with Apple services and 1Password. The update also adds video creation capabilities, new bot characters, and X Spaces support. Additional changes include font customization options and design polish improvements. The features were highlighted in a walkthrough video by creator Matt Palmer on X.
Future Tools · 10-27 17:00 UTC · 同链接合并评分
DeepMind 发布 Genie 4，称其为新一代世界模拟技术。
AI 判断：这是前沿世界模型的发布，视觉演示可能有看点，但缺少普通人可直接使用的功能与价格信息。
展开原始摘录
Research · 6h ago Matt's Pick DeepMind's Genie 4 Brings Real-Time World Simulation With Audio and Advanced Physics DeepMind has announced Genie 4, a natively multimodal world simulation model that generates real-time interactive environments with synchronized audio including music, ambient sounds, and dialogue. Built on Gemini Omni for visual realism and other Gemini models for scene coherence, Genie 4 lets users reshape simulations via natural language prompts in real time. Applications include robot training environments and interactive experiences. It is currently in research preview, with access planned for Google AI Ultra subscribers via Project Genie and Gemini API partners.
deaux · 10-27 16:51 UTC
Ghostty 1.4
AI 判断：前置规则过滤：HN 标题及已抓取内容未命中 AI 主题
展开原始摘录
--- Top Comments ---
[mitchellh]: These are Ghostty GUI release notes, but I want to take a moment to highlight the libghostty progress we've made since Ghostty 1.3. As a reminder, libghostty is the highly portable, embeddable library that provides core terminal emulation, render state, and more. I always felt that a free, high qualty, lightweight, reusable terminal emulation core would benefit the developer world more than any single GUI application. And that is playing out! It is and always was my true purpose with the...
[jaywcarman]: I've tried Ghostty and it's great, but Kitty's hints mode[1] is the one feature keeping me from switching from Kitty to Ghostty. There are quite a few discussions around this and similar features in the Ghosstty-org/ghostty Discussions[2] which I follow. Most of them are references from ghostty-org/ghostty/discussions/2394[3]. Maybe now that Ghostty has finished implementing Kitty graphics hints will be next . [1]: https://sw.kovidgoyal.net/k...
[cassianoleal]: I still don't understand the point of Ghostty, Kitty, Wezterm... Konsole is way faster than any of those on my computer and it has all the usual suspects I'd expect from a modern terminal emulator. The one feature I miss since moving to Linux and not having iTerm2 is tmux -CC integration. None of those offer it. To be clear, I'm not attempting to be dismissive. I really want to understand what it is that they bring.
[eclipticplane]: Man, still does not fix the one bug I care about: being able to quickly `⌘+F, ⌘+V` - the find textbox does not respect keyboard shortcuts: https://github.com/ghostty-org/ghostty/discussions/11159
[ktosobcy]: Why should I use Ghostty over iTerm2? I don't care about fancy fonts, iTerm2 speed (for me) is plenty (just tested against 1.4 -- don't see noticeable difference on MBP M1) and I do like to tweak some options/settings (and was just mildly annoyed when hitting `cmd+,` opened my default text editor...
The Verge：AI（RSS） · 10-27 16:51 UTC
据 The Verge 报道，Anthropic 计划用特殊治理结构，在 IPO 后抵御股东要求其在 AI 安全上妥协的压力。
AI 判断：标题初筛（未依据正文）：这是 IPO 与公司治理层面的新闻，与普通人使用 AI 没有直接关系，也缺少可演示、可复现的画面。
展开原始摘录
Anthropic 计划通过长期受益信托和 Founder LLC 等独特治理结构上市，在公开市场融资的同时防止股东迫使公司削弱安全投入。信托的五名投票受托人可通过 Class T 股票任命七名董事中的四席，七位创始人则以 Founder LLC 掌握 50.1% 投票权。法学专家认为法律层面可隔绝逐利压力，但市场竞争、融资需求和创始人离开都可能侵蚀这些保障。
LangChain 解析自研设计 agent Content Studio 并开源其 harness
LangChain：Blog（RSS） · 10-27 16:45 UTC
LangChain 解析其自研设计 Agent Content Studio，并开源其 harness。
AI 判断：标题初筛（未依据正文）：标题显示这是面向开发者的 Agent 架构与开源 harness，普通人难以直接使用；判断仅基于标题。
展开原始摘录
LangChain 详解自家设计 agent Content Studio 如何在 Slack 中根据一条提示生成视频、幻灯片和社媒图等代码化资产，已累计产出超 3000 件资产、80 多套幻灯片，单次请求模型成本从约 $8 降至约 50 美分。
Classification of Finite Simple Groups Formalized in Lean
pfdietz · 10-27 16:29 UTC
一个研究团队宣称已在 Lean 证明助手中形式化有限单群分类定理，这是数学形式化领域的重大工程进展。
AI 判断：这是面向数学家和形式化验证社区的纯技术进展，与普通人的工作生活基本无关，也缺乏可直观演示的画面。
展开原始摘录
--- Top Comments ---
[mlmonkey]: Out of curiosity: is it possible to take a frontier model and just let it loose on all math textbooks out there, converting each and every proof/theorem/lemma/etc. into Lean? A frontier model should be able to do that, right? And it's a one-time task: do it once and it's there for eternity, available as a library to be used in other proofs. Maybe my understanding of Lean is wrong; I'm thinking of it as a programming language but for Math.
[sebzim4500]: This seems like a very good use of auto-formalization, because in the past mathematicians have incorrectly believed that they were finished with the classification only discover a missing class/sporadic group. The original proof was a collaboration of ~100 mathematicians, if Tao is correct then we should expect to see more such collaborations in the future like we do in e.g. physics.
[vessenes]: Note they found one gap (and filled it) in the previously accepted 1,000+ page proof. So that's pretty cool.
[sleepyams]: Somehow the classification of finite simple groups makes me feel uneasy as a mathematician, but I suppose this helps quell that feeling. https://en.wikipedia.org/wiki/Monster_group
[compacct27]: Groups were my gateway drug into abstract algebra. A friend clued me into symmetry being everywhere and Groups are how symmetry is formalized, and it’s been a crazy journey understanding all that. The frontier models can tie those abstract concepts to plain English examples and answer questions decently well, it’s never been a better time to pick up the field solo and see what it has to offer us
Amazon Is Gutting Its AI Division After Sustained Failure
dpflan · 10-27 16:25 UTC
有报道称，亚马逊在 Nova 等 AI 业务持续受挫后，正大幅收缩其 AI 部门。
AI 判断：这是大公司 AI 战略受挫的行业新闻，对普通人的直接实用价值有限，且报道细节仍需核实。
展开原始摘录
--- Top Comments ---
[atleastoptimal]: Either you are getting the smartest engineers, or offer the most freedom to pursue interesting goals that aren't on the frontier. Amazon has neither so they naturally aren't going anywhere in that domain.
[futuraperdita]: Anyone that's been in the Amazon offices in the last few years is not likely to have witnessed a culture that fosters lab-grade innovation.
[Foxhuls]: This article was published on July 29th. I’m not saying it may not still be interesting or whatever but I thought that was worth pointing out.
[kazinator]: I just played with Nova after seeing this submission. I gave it some real problems for which I used Gemini. It markedly "dumb" in comparison, losing sight of facts that I presented in the first prompt of the chat, and other subsequently introduced, and its responses contain repetitions of the same point-form advice. It's like GPT-3 level or something.
[jdw64]: When will GPU servers become affordable
New 14‑inch MacBook Pro with M6
beex · 10-27 16:08 UTC
苹果官网新闻稿宣布推出搭载 M6 芯片的新款 14 英寸 MacBook Pro，属于常规硬件迭代，未提及具体的 AI 新能力或使用场景。
AI 判断：这是苹果官网硬件新品的宣传稿，缺少可演示、可验证的 AI 功能与实用信息，对普通人的 AI 使用价值有限。
展开原始摘录
--- Top Comments ---
[xlayn]: I want to propose a called AI Index, if you can do say 1000t/s on processing and 50t/s on generation on a min 25B model you can call your device AI9 and you have AI7 and AI5 and AI3 with 3 being 100t/s processing and 10t/s generation on a min 10B model.
Values above are on a 100k context size minimum.
OFFLINE. So any effing claim of being an AI device doesn't end on the demo of best buy laptop where you remove a dog from a photo.... Or that you have "better noise...
[swiftcoder]: I have to say, I really dislike the thing where the base M(n+1) is always in the market alongside M(n) Pro/Max. I appreciate the underlying chips are probably quite distinct, but from a naming standpoint its confusing as hell.
[at1as]: I'm currently on a marathon to keep my M1 Pro running until the M7 Pro/Max is available. Not sure if I'll make it, but I'm not eager to pay $6000+ for a decent spec dev computer, until I really have to
[steelframe]: I switched entirely over to Asahi Linux on my M1 Macbook Air and installed a third-party 55Wh battery, as the original battery had fallen below 80% capacity. That machine is good to go for another 5 years. There are no features or performance characteristics of any of the subsequent M[2-6] Macbook Air or Pro models that I care about. I'm not really sold on running models on the hardware I have right in front of me. So long as I can rent an RTX PRO 4000 Blackwell at about half a buck per ...
[grd_dev]: I have been a fan of these newer macbooks. They are just pushing the boundaries of a good laptop. While I understand the use, I hope someday my AAA games are also as supported in MacBooks as my gaming laptop.
Google DeepMind：Blog（RSS） · 10-27 16:01 UTC · 同链接合并评分
DeepMind 发布 Genie 4，称其为新一代世界模拟技术。
AI 判断：这是前沿世界模型的发布，视觉演示可能有看点，但缺少普通人可直接使用的功能与价格信息。
展开原始摘录
Google DeepMind 发布世界模型 Genie 4，主打实时交互模拟、更强照片级真实感，并首次在实时世界模型中原生生成音频，包括环境声和可实时交互的人物对话。
OpenAI & Oracle Expand Partnership, Bringing Business Apps and AI Agents to ChatGPT
Future Tools · 10-27 16:00 UTC
OpenAI 与 Oracle 扩大合作，计划把 Oracle 企业应用和数据接入 ChatGPT，并推出面向企业工作流、安全与小型企业的 AI 产品。
AI 判断：这是企业级集成新闻，普通用户难以直接使用，也缺少可亲自演示的消费级功能。
展开原始摘录
Company News · 7h ago OpenAI & Oracle Expand Partnership, Bringing Business Apps and AI Agents to ChatGPT OpenAI and Oracle are expanding their partnership to bring Oracle applications and data into ChatGPT and power new Oracle products with OpenAI models. Oracle is developing Oracle AI Workspace to let users complete business workflows inside ChatGPT. New offerings include Oracle AI Security Mesh for finding vulnerabilities, Oracle AI SMB Bundle combining ChatGPT Business and Codex for small businesses, and Oracle App Composer for building apps. Oracle reports 130,000 active ChatGPT users and 95,000 active Codex users.
Replit：Blog（RSS） · 10-27 15:30 UTC
Replit 宣布其 Cloud Agents 允许包括非技术人员在内的成员向企业代码库提交经过测试的 PR。
AI 判断：标题初筛（未依据正文）：仅凭标题判断，这属于面向企业开发流程的开发者工具功能，普通用户难以直接使用或演示，因此按技术类低分处理。
展开原始摘录
Replit 推出 Cloud Agents，面向 Pro 与 Enterprise 客户开放，让设计、客服等非工程人员用自然语言向公司核心代码库提交经测试的 pull request。
Anthropic trained Claude on illegally obtained Stanford professors' books
car · 10-27 15:28 UTC
斯坦福日报报道称，Anthropic 在训练 Claude 时使用了盗版获取的斯坦福教授著作，引发对 AI 训练数据来源合法性的讨论。
AI 判断：这是关于 AI 训练数据版权与伦理的及时争议，适合做有事实依据的观点解读，但缺少可直接演示的普通用户价值。
展开原始摘录
--- Top Comments ---
[piltdownman]: And I presume the Anthropic CEO will be indicted on 13 felony counts under the Computer Fraud and Abuse Act and for wire fraud. Like Aaron Swartz, right? Right? N.B. JSTOR explicitly settled with Swartz, suffered no financial harm, and begged the federal government to drop the case. The government refused and prosecuted Swartz anyway. Anthropic settled for $1.5 billion primarily because of the theft and storage of the files, but this was somehow a Civil issue...
[rafram]: > “To put me to work that way, without remittance, is a form of slave labor,” said Wolff, professor emeritus of English and winner of the National Medal of Arts. “They’re using my work without paying for it. As time goes on, they’re going to make the lives of writers very difficult if there’s no fruit to their labor.” This is obviously hyperbolic, but it was a very poor choice of words. It also relies on an assumption - that AI will become a substitute for buying books - that I think is ...
[awakeasleep]: That professor saying that the use of work he already produced amounts to "slave labor"... copyright infringement is substantially different than slavery.
[zelphirkalt]: In a just world they would have to pull all models, for which this is the case, have auditors walking in left and right, supervising their training of the next version, to make sure they only train on material, that they are allowed to train on, and meticulously have them document every little corpus of data they use and its sources. In reality we won't see this behavior stop, until Anthropic is hit with a huge penalty sledgehammer that threatens its existence. This is a gold rush and wi...
[kenmacd]: The title is (I believe intentionally) inaccurate. Better would be: Anthropic trained on illegally acquired Stanford professors' books The article does not say the training was illegal (although does discuss the professors push for a licensing model).
8 examples of Gemini 4 Argon in action
Google Official Blog · 10-27 15:00 UTC
该文标题称将展示 Gemini 4 Argon 的 8 个实际应用示例。
AI 判断：标题初筛（未依据正文）：标题指向可演示的消费级模型应用，但仅有标题，具体功能、可用性与可复现性未知。
展开原始摘录
Collage of various uses of Gemini 4 Argon
Retell uses OpenAI models to help power 100M monthly calls
OpenAI News · 10-27 15:00 UTC
Retell 利用 OpenAI 模型来支撑其每月 1 亿次通话。
AI 判断：标题初筛（未依据正文）：标题显示这是企业语音 AI 平台的客户案例，偏开发者与企业集成，普通用户难以直接使用。
展开原始摘录
GPT-6 Astra tops Retell’s voice agent benchmark, while Codex helps it reach a $1.6M annual revenue run rate per employee.
Is Anthropic the best business in the world?
Ben's Bites · 10-27 14:19 UTC
该文探讨 Anthropic 是否堪称全球最佳商业模式。
AI 判断：标题初筛（未依据正文）：标题仅提出商业价值疑问，缺乏普通人可用的功能或可验证信息，属观点讨论。
展开原始摘录
Lots of visuals to use and learn with.
helterskelter2 · 10-27 14:06 UTC
US Government is hiding a secret court ruling about how it spies on Americans
AI 判断：前置规则过滤：HN 标题及已抓取内容未命中 AI 主题
展开原始摘录
--- Top Comments ---
[edot]: "Wyden has also warned about a classified interpretation of Section 702 that he says multiple administrations have relied on and refused to declassify. “When it is eventually declassified,” he said in March, “the American people are going to be stunned that it took so long and that Congress had been debating this authority with insufficient information.”" Spying on American citizens is a bipartisan point of agreement.
[TheCoelacanth]: Secret courts are anathema to democracy and need to be abolished.
[V__]: I will always be baffled by the dichotomy of "the US has the most freedom" and "no democratic country spies more on its citizens".
[silverFork]: How it is done, everybody knows how.. backdoors in servers, back door in communication equipment, decryption method for of all that goes through fiber in entire world.. Massive data storage to store everything. How is not an issue; every node and communication, hardware and software has a backdoor. Do they have the right to tap into world communication as well as their own is the issue? They are not challenged by apposing parties and their own population or people have given up. Perhaps...
[giancarlostoro]: Isn't that what all the hype about National Security Letters (and related events from those years) were all about? I never thought any of that went away, the media just stopped reporting on it. We had Lavabit and Snowden showing us how unhinged these things can get.
Here be monsters: What do you still need to know when agents do the work?
adrq · 10-27 13:23 UTC
一篇从开发者视角反思智能体接手工作后仍存在哪些风险和盲区的文章。
AI 判断：内容偏智能体工程与可靠性讨论，缺少普通人可亲自演示或直接使用的功能。
展开原始摘录
--- Top Comments ---
[mrothroc]: This matches my own experience, pre-AI. As I've moved from IC to having a team over the years, more code was being written than I could possibly understand at the depth I would have if I'd written it myself. The skill I learned is exactly as you describe: understand the system. For me, this means "boxes and lines", the components and how they interact. I've encouraged developers on my team to make sequence diagrams part of their deliverable, because it helps both of u...
[trjordan]: I like the analogy. It feels right. But I don't think it actually is, because it leans hard on the _feeling_ of failure, vs. the actual failure. The failure that actually matters here is product failure. Angry users. Security incidents. Downtime. Broken workflows. Stressed and uncomfortable developers have, historically, been a leading indicator of product failure. Software is complex and when devs can't handle the breadth of ownership + the depth of expertise at the same time, it m...
[dgrcode]: > They may have been specialists in certain things, but what was most impressive was their ability to draw upon and weave together seemingly disparate concepts from different disciplines and use that knowledge to solve problems. That's a nice way to put it.
[smallerfish]: Site is down, but one thing in particular that needs close attention: db schema. I've found that LLMs make awful normalization decisions; you can limit the fallout somewhat by modularizing database access, bit it still inevitably bleeds over into the object model etc. On AI heavy projects I tend to pause feature development every few days for code & schema review, and subsequent cleanups.
[LewisOrton]: I dont know if i fully agree to it. ppl are still trying to find the aspects that AI do not handle (or do very well) yet, to define the irreplaceable value of human (humanity?), without realizing that these abilities/capabilities are within the expanding range of AI soon if not tomorrow. Why would one assume that (99.99% of) human can solve the issues that AI can not? should we stop treating the idea that humans must possess some capability AI can never replace as the starting point for ...
搭载 M6 芯片的新款 iMac
2.85New iMac with M6
notfried · 10-27 13:02 UTC
苹果官网出现搭载 M6 芯片的新款 iMac 产品页面，内容为硬件产品介绍，目前未见可核实的 AI 功能细节。
AI 判断：这是苹果硬件产品页而非 AI 新闻，缺少可演示的 AI 用法和具体事实信息，与账号定位不符。
展开原始摘录
--- Top Comments ---
[chrisguilbeau]: I've bought several iMacs (2008, 2012, 2014, 2020) Anyway, it is frustrating once they aren't supported anymore. I think they are overbuilt, they are built to last forever but designed to be obsolete in 7 years. ChromeFlex has saved a few from recycling at our house (A chromebook with a retina display seems like such a waste) I suggest spending a bit more and getting a mac mini and a studio display instead (Equally well built but can be used almost forever). If you could only (easil...
[driverdan]: I don't understand why someone would buy this. It's $1700 for the same computer as a $900 Mac mini. The antiquated 24" monitor, keyboard, and mouse are not worth the extra $800. You can buy a very nice 32" monitor and better input devices for less than that.
[JaggerJo]: I wish they would make big displays again. Like 32” big 6K. 24” or even 27” feels like such a downgrade. Also please make the iMacs usable as external monitors agains. Would buy one as a “livingroom computer” instantly if it was also possible to occasionally use it as a docking station for an air..
[Xeoncross]: We're getting close to local AI being really useful. Before the M1 is dead we might have already seen a huge shift. In my mind, there are two types of macs: those that have the bandwidth to run local models (Max & Ultra) and those that don't (Base & Pro chips). The generation doesn't mater as much to me as the memory bandwidth. When that day comes, I'd rather have a M1 Max 64GB over a M5 Pro 64GB.
[HelloUsername]: I'll wait for the M7 iMac with OLED and then never buy it
OpenAI 请外部专家核查其 AI 在数学、计算机科学和物理学领域产出的证明
The Decoder：AI News（RSS） · 10-27 12:37 UTC
据标题，OpenAI 邀请包括一位诺贝尔奖得主在内的外部专家，核查其 AI 在数学、计算机科学和物理学中生成的证明。
AI 判断：标题初筛（未依据正文）：标题仅显示科研证明核查，面向普通人缺乏可直接使用或演示的场景，属偏技术类信息。
展开原始摘录
The Decoder 报道 OpenAI 请外部专家担任"shepherds"核查其 AI 在计算机科学、数学和物理学领域的三项证明结果，并将部分证明翻译成 Lean 供计算机逐步验证。
What shoppers choose once they’ve tried a retail AI agent
ElevenLabs Blog · 10-27 12:00 UTC
ElevenLabs 博客讨论了购物者在试用零售 AI 智能体之后的选择与偏好。
AI 判断：标题初筛（未依据正文）：仅从标题看，这是厂商博客对零售 AI 客服使用体验的讨论，缺少普通人可复现的演示或具体使用信息。
展开原始摘录
Retail is becoming conversational. People consistently choose speaking over typing on every shopping task. Dive into the data in this article.
f311a · 10-27 11:17 UTC
Python's experimental JIT to be removed
AI 判断：前置规则过滤：HN 标题及已抓取内容未命中 AI 主题
展开原始摘录
--- Top Comments ---
[adrian17]: Unfortunate. I've been following the work on it occasionally. On one hand, it seemed like JIT work was really progressing in high gear until SC requested a new PEP. Also unfortunate, as it seemed like part of the hope for the JIT was that it'd counter the performance penalty from free threading. On the other, it did feel (to me) like with all the fancy (and definitely cool) optimizations it got (automatic refcount operation removal, new type-based specializations, inplace math opera...
[rich_sasha]: Here’s an idea instead (maybe it already exists). Allow a decorator for writing inline Cython. It solves, I think, 90% of the problem. On first run, or eagerly, compile the decorated functions. Basically, Cython is the JIT, and user enables it explicitly. This way I can easily write inlined functions with proper C speed, in a Python esque language. Reasons why I rarely use Cython without this: - unmodified Python code running in Cython is often faster, but not that much. 50% or so is my rule ...
[PathOfEclipse]: How many Python JIT experiments have failed? I find it interesting because Ruby is probably harder to optimize than Python but Ruby has had a JIT for a long time. Maybe base python is more optimized than base Ruby so the JIT is incrementally not changing much? Also, I'd expect javascript to be harder to optimize but it has JITs. That could just be a funding issue, though. There's probably much more money and incentive to optimize Javascript.
[xx_ns]: It's unfortunate that a technically exciting new feature is being killed partly because of what is essentially bureaucracy (and the lack of the developer's personal bandwidth). On the other hand, it's pretty safe to say that the strict adherence to the PEP policy and all of the bureaucracy that comes with it is what makes Python a great language, for better or worse. Not to say that the technical objections were without merit, they likely very much weren't, and the debugge...
[graemep]: If you look at the reasons for this it seems like a good decision: https://discuss.python.org/t/pep-836-jit-go-brrr-the-path-to... There are already many way of addressing Python performance (leaving aside whether Python is a good choice if performance matters) such as Cython, Numba, Pypy and C extensions.
维基百科对AI生成内容的清理
5.75Wikipedia's Cleanup of AI-Generated Content
rbanffy · 10-27 11:08 UTC
IEEE Spectrum 报道维基百科志愿者通过 WikiProject AI Cleanup 识别并清理条目中由 AI 生成的劣质与失实内容。
AI 判断：贴近普通人获取可靠信息的需求，适合做 AI 内容污染与信息可信度的趋势解读，但缺少可直接上手演示的消费级功能。
展开原始摘录
--- Top Comments ---
[grd_dev]: Almost all of these websites or platforms are doing these massive cleanup to remove ai content from them. Well appreciated.
[CPUHall]: Its much appreciated. Reminded me of runescape (a game) wiki devs blogpost, its an interesting read: https://oldschool.runescape.wiki/w/User:Cook_Me_Plox/OSRS_Wi...
[liveoneggs]: I attempted to fix some AI-generated content that I found on wikipedia recently - it had broken formatting, confusing + made-up text, and links on the wrong keywords; just generally very low quality edits. I made my edits and they were immediately reverted by wikipedia's own AI bot. I gave up.
[BatmansMom]: Wow what a nuanced take from Wikipedia. Despite taking such strong efforts to protect human input, still admitting it has a place on the platform: > Yesterday, AI was something to clean up. Today, it’s a tool to help us with the cleanup. Tomorrow, it will be the air we breathe.
[its-summertime]: There was an article, can't remember which, which had an edit that was almost all AI contributions, and added very very little value to the article itself. Looked into the editor, they were participating in one of Wikipedia's school-related programs. I think it makes sense that going out of their way to get contributors who do not care about a topic to contribute, would have this result in part.
The AI bubble meets Anthropic’s real numbers
The Rundown AI · 10-27 10:00 UTC
文章讨论AI泡沫与Anthropic公布的真实数据之间的对照。
AI 判断：标题初筛（未依据正文）：仅从标题看，这是关于AI泡沫与Anthropic真实数据的行业解读，缺少可直接演示的普通人用法。
展开原始摘录
PLUS: Turn sales call notes into follow-up emails with AI
三十亿美元加码!英伟达与Meta联手重塑AI智能体新纪元
AIBase 中文 · 10-27 08:09 UTC
英伟达与Meta宣布深化Muse系列模型及AI智能体基础设施合作，英伟达向Meta投资30亿美元。
AI 判断：这是企业级算力与智能体基础设施合作，缺少普通人可直接使用的新功能或可演示效果。
展开原始摘录
AI资讯 AI新闻资讯 正文 三十亿美元加码!英伟达与Meta联手重塑AI智能体新纪元 发布于AI新闻资讯 发布时间 : 2026年10月27号 16:09 阅读 : 1 分钟 10月27日，英伟达与Meta宣布建立深度合作伙伴关系，双方将全面扩大在Muse及其系列模型上的技术合作，旨在提升推理与训练性能，以高效响应市场对实用型AI智能体的爆发式需求。 根据合作规划，双方将在技术层面展开多维协同:Meta计划依托英伟达OpenShell参考架构强化AI智能体安全体系，并借助英伟达Sentry实现跨结构化、非结构化与图数据全类型场景的智能体行为全链路监控。同时，Meta将进一步扩大部署英伟达Vera Rubin系统、Spectrum-X以太网网络及BlueField-4，全面赋能底层AI工作负载的运行性能与扩展上限。此外，英伟达宣布向Meta的股权发行投资30亿美元。 此次合作标志着算力巨头与大模型领军企业在AI基础设施与智能体安全领域的进一步深度绑定。随着Vera Rubin系统的规模化部署以及Muse系列模型推理效能的升级，双方正加速推动AI智能体从技术验证迈向工业级实用落地，或将重塑下一代人工智能基础设施的竞争格局。 相关推荐 纯国产算力跑出星火X2.6：刘庆峰直言“小米加步枪”硬刚封锁，端侧模型成破局关键 10月22日，在2026科大讯飞全球1024开发者节大会上，刘庆峰谈及国产算力现状与破局之道。他指出，国产算力与英伟达仍有“将近两代”差距，主要集中在显存和带宽；面对百万上下文的代码与智能体任务，常遇“装不下、训不动”困境。为此，讯飞发布端云一体“讯飞星火X2.6”寻求应对。 2026年10月27号 18:03 151.6k 马斯克明年扫货 2 万套英伟达机架，成 Vera Rubin 最大买家 广发证券分析师Jeff Pu预计，SpaceX将在2027年成为英伟达Vera Rubin AI计算机架最大买家。它原计划采购3万套NVL72机架，现预期仅得2万套，但按明年需求仍是新一代系统最大单一采购方。Jeff Pu预计，Vera Rubin NVL72全球规模2026年底达7.3万套，2027年底增至9.6万套，SpaceX的2万套约占明年总量20%。 2026年10月27号 17:14 143.4k 苹果 iOS 27.2 给 AI 图片打上隐身水印 MacRumors称，苹果已在iOS 27.2中悄悄接入谷歌SynthID技术。今后“图乐园”生成的图片、经Apple智能编辑的部分照片，都会加入肉眼不可见的隐藏水印。SynthID在像素层面嵌入标记，用于识别AI生成内容，号称不损画质，并能抵抗裁剪、滤镜、压缩等后期处理。报道还提及在“照片”App中测试相关编辑功能。 2026年10月27号 16:29 162.1k 参数直逼2.8T、支持百万上下文：月之暗面发布开源旗舰大模型K3.1 Pro 月之暗面正式发布全新开源旗舰大模型K3.1Pro。该模型拥有高达2.8T参数规模，支持百万（1M）Token超长上下文，刷新行业对国产大模型性能与美学表现的认知上限。官方确认，完整模型权重预计11月4日前全面释出，供全球开发者与研究人员下载使用，开放节奏与生态建设备受关注。 2026年10月27号 16:28 190.1k 即梦AI上线“短剧工作室”，一站式打通全流程AI创作 10月26日，即梦AI上线“短剧工作室”。该功能定位短剧创作一站式AI工具，把剧本整理、角色与场景管理、分镜生成、视频制作及后期调整等全流程整合至同一平台，免去创作者多平台切换的繁琐。部分功能还整合小云雀短剧Agent同款能力，助即梦发挥优势，打造更全能平台。 2026年10月27号 15:34 174.2k 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
Ai2 发布 Old Ideas, Novel Problems 研究与 benchmark，揭示 LLM 新颖性裁判不稳定
Ai2 / Allen Institute for AI（RSS） · 10-27 08:00 UTC
Ai2 发布了一项名为 Old Ideas, Novel Problems 的研究与基准，指出 LLM 在判断内容新颖性时表现不稳定。
AI 判断：标题初筛（未依据正文）：标题显示这是面向研究者的基准研究，与普通人可直接使用的功能关联较弱。
展开原始摘录
Ai2 发布研究 Old Ideas, Novel Problems、配套 benchmark、数据、代码和自动化流水线，测试 LLM 作为研究想法新颖性裁判的可靠性。
即梦AI上线“短剧工作室”，一站式打通全流程AI创作
AIBase 中文 · 10-27 07:34 UTC
即梦AI上线“短剧工作室”，将剧本诊断、角色与场景资产、分镜生成和视频制作整合到一个平台，并支持10万字剧本与100集连续生成。
AI 判断：这是面向短剧创作者的一站式实用工具，可演示剧本诊断、分镜和资产管理效果，但文中日期标注为2026年，时效性需核实。
展开原始摘录
AI资讯 AI新闻资讯 正文 即梦AI上线“短剧工作室”，一站式打通全流程AI创作 发布于AI新闻资讯 发布时间 : 2026年10月27号 15:34 阅读 : 1 分钟 10月26日，即梦AI正式上线全新功能“短剧工作室”。作为一款面向短剧创作场景推出的一站式AI工具，它将剧本整理、角色与场景管理、分镜生成、视频制作及后期调整等全流程深度整合到同一个平台中，彻底告别了以往创作者需要在多个平台之间来回切换的繁琐体验。据悉，该工作室的部分功能还整合了小云雀短剧Agent的同款能力，旨在帮助即梦进一步发挥优势，打造更加全能好用的内容创作平台。 在核心能力方面，“短剧工作室”搭载了即梦首个专业剧本模型DreamScript。该模型专门面向短剧场景进行训练与优化，完整覆盖了从剧本生成到深度诊断的全链路。用户上传已有剧本后，“剧本医生”功能可以自动诊断前三集的主线，检查主角身份与核心冲突是否清晰;“台词体检”功能则能对照角色小传逐句诊断改编。同时，该产品支持10万字的长文本上传以及100集连续生成。 在资产管理与画布交互上，“短剧工作室”同样迎来了深度优化。上传剧本后，系统能够自动解析全剧资产，将主次人物、场景、道具甚至音色服装精准识别并绑定至资产库。画布与分镜故事板共享同一套资产，支持一键调用与历史版本回溯，做到一处修改、全局同步，后续创作同系列内容时可直接复用。此外，针对庞大的素材堆叠场景，其画布交互支持上千个素材堆叠下依然跟手流畅、缩放顺滑，撤销与重做无卡顿，图片场景下的操作流畅度达到了同类产品的十余倍，节点规模扩展至1万个仍能保持快速响应。 为了更好地适配专业工作室与机构团队的多人协作需求，“短剧工作室”同步推出了团队版管理能力。该功能支持多账号资产管理与统一积分管理，方便团队负责人统筹项目资源并合理分配创作额度。随着此次全新工具的上线与双十一限时优惠活动的推出，即梦正持续通过整合经过验证的AI创作能力，让各类创作者在熟悉的生态中高效完成从灵感到成片的完整全流程。 相关推荐 纯国产算力跑出星火X2.6：刘庆峰直言“小米加步枪”硬刚封锁，端侧模型成破局关键 10月22日，在2026科大讯飞全球1024开发者节大会上，刘庆峰谈及国产算力现状与破局之道。他指出，国产算力与英伟达仍有“将近两代”差距，主要集中在显存和带宽；面对百万上下文的代码与智能体任务，常遇“装不下、训不动”困境。为此，讯飞发布端云一体“讯飞星火X2.6”寻求应对。 2026年10月27号 18:03 151.6k 马斯克明年扫货 2 万套英伟达机架，成 Vera Rubin 最大买家 广发证券分析师Jeff Pu预计，SpaceX将在2027年成为英伟达Vera Rubin AI计算机架最大买家。它原计划采购3万套NVL72机架，现预期仅得2万套，但按明年需求仍是新一代系统最大单一采购方。Jeff Pu预计，Vera Rubin NVL72全球规模2026年底达7.3万套，2027年底增至9.6万套，SpaceX的2万套约占明年总量20%。 2026年10月27号 17:14 143.4k 苹果 iOS 27.2 给 AI 图片打上隐身水印 MacRumors称，苹果已在iOS 27.2中悄悄接入谷歌SynthID技术。今后“图乐园”生成的图片、经Apple智能编辑的部分照片，都会加入肉眼不可见的隐藏水印。SynthID在像素层面嵌入标记，用于识别AI生成内容，号称不损画质，并能抵抗裁剪、滤镜、压缩等后期处理。报道还提及在“照片”App中测试相关编辑功能。 2026年10月27号 16:29 162.1k 参数直逼2.8T、支持百万上下文：月之暗面发布开源旗舰大模型K3.1 Pro 月之暗面正式发布全新开源旗舰大模型K3.1Pro。该模型拥有高达2.8T参数规模，支持百万（1M）Token超长上下文，刷新行业对国产大模型性能与美学表现的认知上限。官方确认，完整模型权重预计11月4日前全面释出，供全球开发者与研究人员下载使用，开放节奏与生态建设备受关注。 2026年10月27号 16:28 190.1k 三十亿美元加码!英伟达与Meta联手重塑AI智能体新纪元 10月27日，英伟达与Meta宣布建立深度合作，全面扩大Muse及其系列模型技术合作，提升推理与训练性能，以满足市场对实用型AI智能体的爆发式需求。Meta将依托英伟达OpenShell参考架构强化AI智能体安全体系，并借助Sentry对结构化、非结构化和图数据等全类型场景，实现智能体行为全链路监控。 2026年10月27号 16:09 152.7k 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
Latent Space 解读 Anthropic 提交 S-1：拟以代码 AN 上市纳斯达克，H1 2026 营收 16.3B 美元
Latent Space（RSS） · 10-27 07:24 UTC
Latent Space 解读 Anthropic 的 S-1 文件，称其计划以代码 AN 在纳斯达克上市，并提及 2026 年上半年营收 163 亿美元。
AI 判断：标题初筛（未依据正文）：标题仅涉及 IPO 与营收信息，属于融资财务新闻，缺少普通人可用的 AI 功能或可演示内容。
展开原始摘录
Latent Space 汇总解读 Anthropic 公开提交的 S-1 文件，公司计划以代码 AN 上市纳斯达克，营收从 2023 年的 41M 美元增至 H1 2026 的 16.3B 美元，Q2 营收 11.6B 美元同比增长 1,368%，营业利润 974M 美元，H1 2026 毛利率达 85%。
OpenAI News · 10-27 07:00 UTC · 同链接合并评分
OpenAI 与 Oracle 扩大合作，计划把 Oracle 企业应用和数据接入 ChatGPT，并推出面向企业工作流、安全与小型企业的 AI 产品。
AI 判断：这是企业级集成新闻，普通用户难以直接使用，也缺少可亲自演示的消费级功能。
展开原始摘录
OpenAI and Oracle are expanding their partnership to bring business applications and AI agents into ChatGPT and power new Oracle offerings.
OpenAI：官网动态（RSS · 排除企业/客户案例） · 10-27 07:00 UTC · 同链接合并评分
OpenAI 与 Oracle 扩大合作，计划把 Oracle 企业应用和数据接入 ChatGPT，并推出面向企业工作流、安全与小型企业的 AI 产品。
AI 判断：这是企业级集成新闻，普通用户难以直接使用，也缺少可亲自演示的消费级功能。
展开原始摘录
OpenAI 宣布扩大与 Oracle 的合作，将 Oracle 应用和数据接入 ChatGPT，并用 OpenAI 模型驱动 Oracle 新产品。
王的猜想，被 AI 一夜通关
4.25爱范儿 · 10-27 04:57 UTC
标题称一个被称为“王的猜想”的问题被 AI 在一夜之间攻克，但仅凭标题无法确认具体内容和可验证性。
AI 判断：标题初筛（未依据正文）：标题只给出 AI 破解数学猜想的悬念，缺少普通人可用场景和可核实细节，适合谨慎处理。
展开原始摘录
AI 已经开始陆续交卷了，但人类可能连阅卷都快来不及了。
#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。
Nikunj Kothari: Maybe I should just change my job.. https://t.co/kHDC0z48LF
Nikunj Kothari · 10-27 03:56 UTC
作者只表示也许应该换工作，并附上一个链接，没有说明具体内容。
AI 判断：标题初筛（未依据正文）：仅凭标题看，这只是模糊感叹加链接，缺少可核实信息，难以形成面向普通人的实用教程或明确观点。
展开原始摘录
Maybe I should just change my job.. https://t.co/kHDC0z48LF
Peter Yang: @bot This is probably not a good thing overall
Peter Yang · 10-27 03:52 UTC
这是一条回复 @bot 的推文，作者认为整体上这可能不是好事。
AI 判断：标题初筛（未依据正文）：标题只有模糊个人评论，缺少可核实事件和普通人可用的具体信息，难以做成教程或演示。
展开原始摘录
@bot This is probably not a good thing overall
Peter Yang: In the near future, instead of reading or doing anything online, people are just going to tag...
Peter Yang · 10-27 03:48 UTC
一条推测未来人们上网将变成“打标签”而非阅读或做事的不完整观点，没有给出依据或具体场景。
AI 判断：标题初筛（未依据正文）：仅标题且内容截断的推测性观点，缺少可核实事实、可演示效果和普通人可用的具体信息。
展开原始摘录
In the near future, instead of reading or doing anything online, people are just going to tag their agent to do it.
@bot is an early signal of this.
Aditya Agarwal: More and more I am coming to accept that our understanding of the sciences were but limited b...
Aditya Agarwal · 10-27 02:31 UTC
Aditya Agarwal 发帖表示，他越来越接受人类对科学的理解是有限的。
AI 判断：标题初筛（未依据正文）：依据标题仅见一句缺乏 AI 具体事实的哲学感慨，难以转化为普通人可用的教程或可核验演示。
展开原始摘录
More and more I am coming to accept that our understanding of the sciences were but limited by our desire to have solutions that we could understand.
Peter Yang: This is the stuff that I’m excited about not the 9th personal agent to manage my emails https...
Peter Yang · 10-27 02:20 UTC
这条帖子仅表示作者对某些AI方向感到兴奋，并明确不是又一个邮件管理个人智能体，但未说明具体内容。
AI 判断：标题初筛（未依据正文）：标题只是模糊预告，缺少可演示功能或事实细节，难以转化为普通人教程或可靠观点。
展开原始摘录
This is the stuff that I’m excited about not the 9th personal agent to manage my emails https://t.co/HB5ssCfYjq
Amjad Masad: OpenAI asked three scientific communities how to release its discoveries and got three differ...
Amjad Masad · 10-27 01:23 UTC
Amjad Masad 表示，OpenAI 就如何发布其发现咨询了三个科学社区，并得到三种不同回应。
AI 判断：标题初筛（未依据正文）：仅据标题，这是一条被截断的观点引述，既无可演示功能也缺少普通人可用信息，受众和实用性偏低。
展开原始摘录
OpenAI asked three scientific communities how to release its discoveries and got three different answers:
Mathematicians: “Just post it. Who gives a shit anymore.”
Computer scientists: “Open source it, but also pay for our conference.”
Physicists: “Leave it to us, nerds.” https://t.co/5cpEOIADpY
Peter Steinberger: I spent some time on the weekend to test bun's AOT compiler and results are nothing short of...
Peter Steinberger · 10-27 00:40 UTC
该帖标题称作者周末测试了 Bun 的 AOT 编译器，并表示结果令人惊讶，但正文结论未知。
AI 判断：标题初筛（未依据正文）：这是开发者工具链的编译器测试，属于纯技术内容，普通用户难以使用或演示，仅凭标题也无法核实结论。
展开原始摘录
I spent some time on the weekend to test bun's AOT compiler and results are nothing short of amazing. Performance up 2-3x, memory use less than half. https://t.co/2yobmkrUUa
Peter Steinberger: Banger. https://t.co/MzQY2HvarM
Peter Steinberger · 10-27 00:38 UTC
Peter Steinberger 仅发布了“Banger”一词和一个短链接，未说明具体内容。
AI 判断：标题初筛（未依据正文）：标题只有含糊感叹与短链接，缺少可核实信息，无法形成教程或实质观点。
展开原始摘录
Banger. https://t.co/MzQY2HvarM
X：卡兹克 (@Khazix0918) · 10-27 00:34 UTC
据标题，Anthropic 招股书被拆解出 11 个细节，其 IPO 计划募资约 1000 亿美元。
AI 判断：标题初筛（未依据正文）：纯融资与 IPO 新闻，对普通人的实用价值低，仅适合作为 AI 行业趋势谈资。
展开原始摘录
作者总结 Anthropic IPO 招股书的 11 个细节，计划募资约 1000 亿美元、估值或达 1.5万亿～2万亿。收入五个季度涨 30 倍，今年二季度经营利润首次转正 9.74 亿美元，但上半年净亏 807 亿美元，主要来自可转换票据公允价值变动；调整后净利润为正 26.71 亿美元。
公众号：数字生命卡兹克 · 10-27 00:05 UTC
一篇文章解读 Anthropic IPO 招股书的 11 个细节，涉及千亿募资、807 亿美元亏损和 7 美元控股权。
AI 判断：标题初筛（未依据正文）：仅凭标题判断，这属于融资与 IPO 财务解读，普通人难以直接使用；并且标题中的巨额数字缺乏核实，需要谨慎对待。
展开原始摘录
作者梳理了 Anthropic 提交的 IPO 招股书中的11个细节。
How Check Point accelerates vulnerability research with Daybreak
OpenAI News · 10-27 00:00 UTC
标题称 Check Point 使用 Daybreak 来加速漏洞研究工作。
AI 判断：标题初筛（未依据正文）：仅凭标题判断，这是面向安全研究人员的开发/技术类内容，普通用户相关性很低。
展开原始摘录
With OpenAI Daybreak, Check Point speeds up the research behind new security protections, enabling its analysts to turn vulnerability intelligence into customer defenses faster.
Expanding our partnership with GSA to bring Claude to all levels of government
Anthropic News · 10-27 00:00 UTC
Anthropic扩大与美国总务管理局的合作，为政府机构购买Claude企业版、政府版和API提供40%折扣，并推出专门的支持计划。
AI 判断：这是面向政府机构的采购折扣与落地支持，普通个人用户既无法直接使用，也缺乏可演示的新能力。
展开原始摘录
We are expanding our partnership with the General Services Administration (GSA) through a new agreement that gives government partners a 40% discount on Claude Enterprise, Claude for Government, and the Claude API. For the first time, the discount applies across every product surface, including when agencies buy from us directly today and through cloud providers in the coming months. The discount also extends to state, local, tribal and territorial governments. We are also launching a dedicated program to help agencies implement and make the most of their Claude use. These programs are commonly leveraged by our private sector customers and we’re looking forward to bringing those best practices to our public sector partners in the coming weeks. Approximately 500 US federal, state, and local government organizations now use Claude. Weekly Claude for Government active users are up about eightfold since February, and usage across federal civilian agencies has grown about 20-fold in a year. Those organizations run some of the most important systems in the country, usually on tight budgets and with long procurement cycles. This agreement is built to support that work, so they can apply Claude to hard problems without a large upfront commitment. The discounts offered through the agreement are estimated to save the government $1 billion. Agencies are not bound by a minimum spend commitment and will be able to pay for the usage they consume. How Claude is helping public servants get more done Government agencies are using Claude to solve some of their most important challenges. Examples include managing backlogs, records and forms, and modernizing old code. Some of the ways agencies are using Claude include: Utah gave Claude Code to 34 developers across 10 state agencies. 80% had a working environment within 30 minutes, 71% reported saving 20+ hours by the end of the four-week pilot. A US state CIO confirmed their team was able to respond to a ransomware attack in 8 minutes vs. the usual 2-3 days thanks to Claude. A US federal security team used Claude Mythos to detect and patch the vulnerability that would have allowed for the takeover of a critical production system in about two days. A human penetration test the month before had missed it. A cybersecurity professional at a major agency found and patched bugs in critical financial systems. A US state IT developer moved nearly 900 API methods to a new framework in two days. The project had been scoped for a month. A US state IT director used Claude Code to document 12 legacy code repositories, mapping 467 database tables and flagging 27 risks along the way. More than half of the US states, some of the largest localities, and some of the countries largest critical infrastructure operators participated in our SLTT cyber defense program. Meeting government security requirements Agencies can use Claude for their most sensitive unclassified work. Claude for Government, which delivers Claude's coding and agentic work capabilities and provides near-feature parity with commercial offerings at a FedRAMP High authorization level, was made generally available on September 30. Agencies building their own tools can use Claude through Amazon Bedrock or Google Cloud’s Vertex AI. On Bedrock, Claude is authorized for FedRAMP High, with select models authorized for DoD IL4 and IL5 workloads in AWS GovCloud. On Vertex AI, Claude is authorized for FedRAMP High. Claude Enterprise is in process for FedRAMP Moderate authorization through the FedRAMP 20x program, marking the next evolution of how we are bringing the same innovative technology available commercially to the government. Getting started Government agencies can buy Claude at the OneGov discount through GSA's Multiple Award Schedule today. Whether your department or agency already uses Claude or is interested in using it, our Public Sector team will help discuss your options. For more information, contact us here .
GPT-6.1 Terra
1.85Simon Willison · 10-26 23:53 UTC
标题中出现的名称是 GPT-6.1 Terra，具体内容尚未提供。
AI 判断：标题初筛（未依据正文）：仅有标题和来源，无法确认发布、功能或可用性，暂不宜作为普通人选题。
展开原始摘录
GPT-6.1 Terra
It's always interesting to note which aspects of a new model are emphasized in the release announcement.
In the case of GPT-6.1 Terra, OpenAI went with visual reasoning and computer use:
We’re introducing
GPT-6.1 Terra
, a fast, efficient model that combines detailed visual reasoning and precise computer use alongside clearer, more natural writing. A significant step up from GPT‑5.6 Terra, it excels at the communication and computer-based tasks that fill the workday, from drafting proposals and interpreting technical drawings to filling out forms and editing videos.
Terra sits in the same position as Sonnet in the Anthropic model collection - not the cheapest or the most expensive, making it harder to decide when to use it.
Computer use is a good example of a task that benefits from a less expensive model, since you might put it to work on automated, repetitive tasks.
OpenAI's Ted Sanders provided some useful advice in
a comment
on Hacker News:
In the case of GPT-6.1 Terra, I'd say Terra is probably better than Astra and Sol at technical communication, similar on vision and computer use (but much faster!), and a bit worse on most other things. For API developers, it's a nice midrange option if you want something cheaper than Sol but a lot smarter than Luna (especially for vision / computer use / communication). For subscribers, it's not a bombshell release - if you're hitting usage limits, maybe it helps you go a bit further, and if you want clearer communication or faster computer use, it can be a nice option to select.
Here are
the pelicans
for the different reasoning levels. This pelican
for "max"
cost
11.3161 cents
and took 4 minutes and 11 seconds:
Here's
a new comparison grid
showing pelicans for every reasoning level of OpenAI's current frontier lineup: GPT-6 Astra, GPT-6.1 Sol, GPT-6.1 Terra, and GPT-6 Luna.
GPT-6.1 Terra is priced at $1/million input tokens and $5/million output - almost half the price of GPT-5.6 Terra and GPT-6.1 Sol, and around 10x the price of GPT-6 Luna.
Via
Hacker News
Tags:
ai
,
openai
,
generative-ai
,
llms
,
llm-pricing
,
pelican-riding-a-bicycle
,
llm-release
,
gpt
Swyx: will you buy $ANT stock at a $2T valuation?
Swyx · 10-26 23:41 UTC
Swyx 以反问形式提出：是否有人愿意在 2 万亿美元估值下买入 $ANT 股票。
AI 判断：标题初筛（未依据正文）：仅凭标题只有一句关于估值与买入的反问，缺少可核实事实和普通人可用的具体信息。
展开原始摘录
will you buy $ANT stock at a $2T valuation?
IT之家·人工智能 · 10-26 23:37 UTC
据标题，OpenAI 称其 AI 模型在局部光滑猜想、指数时间假设和杨–米尔斯理论三个数学难题上取得突破。
AI 判断：标题初筛（未依据正文）：标题仅给出高深数学突破，普通人的实用价值与可演示性弱，且细节未证实。
展开原始摘录
OpenAI 于 10 月 27 日宣布其 AI 模型在数学、计算机科学和物理学领域取得三项研究突破。数学上给出了所有维度下局部光滑猜想的证明，从而解决挂谷猜想与斯坦因限制猜想；计算机科学上给出推翻指数时间假设（ETH）的证明，暗示相关计算任务可能有更快算法；物理学上提出量子色动力学及标准模型一种统一完备化的严格存在性证明，涉及杨–米尔斯千禧年难题的一部分。
其他 Follow Builders 资讯
中文标题Nikunj Kothari：也许我该直接换个工作……
Follow Builders · X 动态 · Nikunj Kothari · 10月27日 03:56 UTC · 喜欢 0 · 转发 0 · 回复 0
作者只表示也许应该换工作，并附上一个链接，没有说明具体内容。
中文标题Peter Yang：@bot 总的来说，这大概不是什么好事
Follow Builders · X 动态 · Peter Yang · 10月27日 03:52 UTC · 喜欢 7 · 转发 0 · 回复 2
这是一条回复 @bot 的推文，作者认为整体上这可能不是好事。
中文标题Peter Yang：不久的将来，人们上网不再阅读或做任何事，而只是打标签……
Follow Builders · X 动态 · Peter Yang · 10月27日 03:48 UTC · 喜欢 49 · 转发 3 · 回复 37
一条推测未来人们上网将变成“打标签”而非阅读或做事的不完整观点，没有给出依据或具体场景。
中文标题Aditya Agarwal：我越来越接受，我们对科学的理解不过是有限的……
Follow Builders · X 动态 · Aditya Agarwal · 10月27日 02:31 UTC · 喜欢 90 · 转发 4 · 回复 14
Aditya Agarwal 发帖表示，他越来越接受人类对科学的理解是有限的。
中文标题Peter Yang：这是我感到兴奋的东西，而不是第 9 个帮我管理邮件的个人智能体 https...
Follow Builders · X 动态 · Peter Yang · 10月27日 02:20 UTC · 喜欢 25 · 转发 1 · 回复 13
这条帖子仅表示作者对某些 AI 方向感到兴奋，并明确不是又一个邮件管理个人智能体，但未说明具体内容。
中文标题Amjad Masad：OpenAI 曾询问三个科学社区如何发布其发现，却得到三种不同……
Follow Builders · X 动态 · Amjad Masad · 10月27日 01:23 UTC · 喜欢 395 · 转发 20 · 回复 68
Amjad Masad 表示，OpenAI 就如何发布其发现咨询了三个科学社区，并得到三种不同回应。
中文标题Peter Steinberger：周末花时间测试了 bun 的 AOT 编译器，结果非常……
Follow Builders · X 动态 · Peter Steinberger · 10月27日 00:40 UTC · 喜欢 439 · 转发 12 · 回复 49
该帖标题称作者周末测试了 Bun 的 AOT 编译器，并表示结果令人惊讶，但正文结论未知。
中文标题Peter Steinberger：Banger
Follow Builders · X 动态 · Peter Steinberger · 10月27日 00:38 UTC · 喜欢 519 · 转发 42 · 回复 26
Peter Steinberger 仅发布了“Banger”一词和一个短链接，未说明具体内容。
中文标题Swyx：你会在 2 万亿美元估值时买入 $ANT 股票吗？
Follow Builders · X 动态 · Swyx · 10月26日 23:41 UTC · 喜欢 14 · 转发 0 · 回复 51
Swyx 以反问形式提出：是否有人愿意在 2 万亿美元估值下买入 $ANT 股票。