All AI Intelligence

Total 10130 items
Sun·1 items
  • IT之家 · 智能时代T2·Infra·72 pts

    Phison Unveils Smart SSD with Built-in NPU and 64GB Memory

    Original: 群联展示内置 NPU 与 64GB 内存的智能固态硬盘,可直接运行 AI 模型

    At SC26, Phison showed two enterprise SSDs built for AI workloads. One integrates an NPU and 64GB of memory with 130GB/s bandwidth, running PyTorch, vLLM, OpenVINO and llama.cpp directly on the drive, pushing compute to storage and returning only results to the host, which Phison says can cut data movement by 10-100x on certain workloads. The other uses a mixed-media design pairing pSLC cache with QLC flash, limiting cache loss to 4% per failed drive versus 25% in traditional dedicated-cache setups.

Sat·24 items
  • Simon Willison 博客T2·Tools·58 pts

    Graham Dumpleton, creator of mod_wsgi, has launched Muru, an AI-assisted Rust server aimed at providing the best tooling for running ASGI and WSGI Python applications in both development and production. Continuing the approach he used with wrapture, he is publishing an unusually detailed build-in-public account of the project as he develops it.

  • IT之家 · IT业界T2·Industry·38 pts

    HarmonyOS 7 Expands XiaoYi Starlight Voice to More Huawei Devices

    Original: OSTA 新特性放开:华为鸿蒙 HarmonyOS 7 系统更多机型小艺助理上线“小艺星芒”音色

    Huawei has extended its XiaoYi Starlight voice to more HarmonyOS 7 devices such as the Mate 80 and Mate 90, adding pitch and speed controls after debuting the feature on the OSTA O1 Pro. The move is a routine personalization update, with the Starlight line's exclusive AI screen and avatar features still unconfirmed for wider release.

  • IT之家 · IT业界T2·Industry·58 pts

    OpenAI Cuts Chain-of-Thought Extraction Rate to 0.2 Percent

    Original: IT早报 1122:曝其他厂商也考虑推出阔直板形态手机;Anthropic 研究员预测 2027 年底 AI 比肩顶尖专家;华为星芒 O1 Pro 定价 5499 元起引争议...

    OpenAI disclosed anti-distillation safeguards for GPT-6.2 Astra and GPT-6.2 Sol, with black-box red-team tests showing chain-of-thought extraction success dropping from 54% to 0.2% on legacy requests and from 22.9% to 0.9% on new ones. Combined with weekly bans of thousands to tens of thousands of unauthorized distillation accounts, the move signals frontier labs are now actively defending reasoning traces as core assets.

  • IT之家 · 智能时代T2·Industry·72 pts

    Signal Blocks Screenshots by Default to Stop AI Agents Reading Chats

    Original: 加密聊天应用 Signal 默认屏蔽截图,防止 AI 智能体读取用户消息

    Signal now blocks screenshots by default on iOS and Android to prevent system-level AI agents such as Siri AI and Gemini from reading private messages. President Meredith Whittaker framed the move as a defense of end-to-end encryption against expanding on-device AI, while macOS lacks the tooling for equivalent protection.

  • 宝玉 (@dotey)T2·Models·96 pts

    Anthropic Mathematician and Claude Claim Riemann Hypothesis Proof in Lean

    Original: 2026 年 11 月 20 日,Anthropic 的数学家 Levent Alpöge 宣布,他和 Claude 一起证明了黎曼猜想,交出来的是 1700 万行 Lean 代码。

    Anthropic mathematician Levent Alpöge announced that he and Claude have produced a proof of the Riemann Hypothesis, delivered as 17 million lines of Lean code, which the Lean FRO and Mathlib community have jointly certified as passing all verification criteria. It marks the first time AI has contributed to cracking a 160-year-old top-tier open problem at this scale of formal verification. On the same day, a purely human proof draft from Stanford was flagged with issues by AI within hours, offering a sharp contrast.

  • IT之家 · 智能时代T2·Industry·58 pts

    Google to Shut Down Nest App in 2027, Folding Features into Google Home

    Original: 谷歌智能家居应用 Nest 明年停服,功能全面整合进 Google Home

    Google will shut down the standalone Nest app and web portal on January 15, 2027, automatically migrating thermostats, cameras, doorbells and schedules into Google Home, with legacy Nest Aware subscriptions converted to Google Home Premium at no extra cost. On August 2, 2027, Google Assistant will also be retired, with Google Home and Gemini taking over first-party speakers and third-party devices such as Sonos.

  • IT之家 · IT业界T2·Industry·78 pts

    SEC Probes Nvidia Accounting as Shares Fall 48% From October Peak

    Original: 消息称美国证券交易委员会对英伟达潜在会计问题展开调查,股价较 10 月高点腰斩

    The SEC's enforcement staff is examining potential accounting issues at Nvidia, Reuters reported, citing two people familiar with the matter; both the regulator and Nvidia declined to comment. The news, following an earlier Wall Street Journal report, deepened investor anxiety and sent shares down more than 14% to $125.5, roughly 48% below October's $243.4 peak.

  • Rohan Paul (@rohanpaul_ai)T2·Models·78 pts

    UT Austin System Beats Human Champion at Times Crossword Championship

    Original: "the first automated system to reach superhuman performance on British cryptic crosswords."

    A University of Texas at Austin team's automated system, Darmok, solved all seven British cryptic crossword puzzles error-free at the 2026 Times Crossword Championship, beating the human champion by 3 minutes 13 seconds in the Final, the first automated system to reach superhuman performance in this event. Notably, the primary model provider slowed sharply during the Final, and backup routing supplied 17 of the 30 answers, underscoring the value of multi-model redundancy for latency-critical tasks.

  • 宝玉 (@dotey)T2·Tools·38 pts

    Antigravity CLI Team Vows Fixes After Reviewing 500 Feedback Replies

    Original: 很多人吐槽 Antigravity 不好用,但只是吐槽不一定会被看见,不如去给 @shengzheyao 留言,这样有机会被看到并解决问题。

    The Antigravity CLI lead read through 500+ user replies and distilled the main pain points: prompt cache efficiency, latency and stream stability, auto-mode reliability, permission friction across workspaces, progress visibility on long runs, and unified sessions across CLI, desktop, remote, and mobile. The team committed to turning the feedback into real fixes, while users are urged to send complaints directly to the lead rather than venting into the void.

  • Hacker News AIT2·Tools·58 pts

    Open Source PcapAnalyzer Ships 27 Network Fault Detectors

    Original: PcapAnalyzer – self-hosted pcap analyzer with 27 problem detectors

    PcapAnalyzer is an Apache 2.0 open-source, self-hosted analyzer for .pcap/.pcapng captures that runs 27 detectors covering TCP retransmissions and window anomalies, PMTUD black holes, DNS/DHCP/ARP failures, TLS alerts, and tunnels such as IPsec, WireGuard and OpenVPN, each finding paired with a severity and next-step guidance. Captures stay on local infrastructure, making it relevant for security-sensitive network operations; the project is currently in beta.

  • Rohan Paul (@rohanpaul_ai)T2·Models·78 pts

    Frontier Models Can Identify Anonymous Authors From Writing Style Alone

    Original: New University of Maryland paper finds that frontier models now detect AI-written prose as well as t...

    A new University of Maryland paper finds that frontier models, given only a single text and a zero-shot prompt with no tools or candidate lists, detect AI-written prose as well as trained detectors like Pangram while explaining their reasoning. More strikingly, they infer authorship: Claude Fable 5.2 identified the reviewer in 56% of 398 anonymous peer reviews and the account behind 69% of comments from Redditors with over 500 posts a year. The anonymity assumptions behind blind review and pseudonymous forums are now weak.

  • TechCrunch AIT2·Industry·62 pts

    Signal Restricts Mobile Screenshots to Counter AI Agent Surveillance

    Original: Signal restricts mobile screenshots as it confronts ‘existential threat’ of AI surveillance

    Signal announced it is restricting in-app screenshots on Android and iOS by default, following a similar move on Windows. President Meredith Whittaker called operating systems with integrated AI agents an existential threat, arguing their screen-scanning capabilities prevent Signal from independently guaranteeing privacy. She acknowledged the fix is incomplete and can be disabled by users, which would break the app's privacy guarantees.

  • Hacker News AIT2·Tools·58 pts

    Rust GUI Development Accelerates on Surging AI-Driven Interest

    Original: Advice for the next 100 Rust GUIs

    Raph Levien's follow-up essay argues that although Rust GUI adoption has been slow over the past four years with few flagship apps, AI-driven interest is now rapidly accelerating the space. He maintains that Rust remains a compelling path to high-performance cross-platform UI and advises newcomers to build on solid foundations, follow established specs, and collaborate constructively with open source upstreams rather than submitting low-quality PRs.

  • TechCrunch AIT2·Industry·58 pts

    Nvidia reported soaring revenue and its current leadership insists there is no plan for former CEO Jensen Huang to return. Yet Elon Musk and other tech executives are circulating a petition calling for Huang's comeback, Huang says he is prepared to return, and the falling stock price signals Wall Street remains unconvinced.

  • Hacker News AIT2·Tools·58 pts

    Pandex Launches European Law API and MCP Server for AI Agents

    Original: Pandex – Official European Law for AI Agents, over MCP

    Pandex has launched a legal data platform covering EU and seven European countries' legislation and supreme court rulings, offering a REST API and MCP server with keyword, semantic and hybrid search that plugs into Claude, Cursor, GitHub Copilot and other AI tools. It holds roughly 2.8 million official documents, with 33 of 41 institutions complete, and links 37 million citations back to their source text.

  • Rohan Paul (@rohanpaul_ai)T2·Industry·58 pts

    Eric Schmidt Says a Mutual AI Race Pause Is Impossible

    Original: Eric Schmidt explains why a mutual pause of the AI race is impossible:

    Former Google CEO Eric Schmidt argues in an Economist interview that a mutual pause in the US-China AI race is unworkable, since either side has incentives to cheat and UN-style verification is technically infeasible. Modern codebases are largely machine-written, he notes, so no human inspection team can review them line by line, and inspectors cannot access secret facilities including military labs.

  • Rohan Paul (@rohanpaul_ai)T2·Models·88 pts

    Claude Autonomously Finds Algorithm Breaking 3SUM and APSP Limits

    Original: New Columbia and MIT paper shows that an algorithm Claude found beats the long-standing speed limits...

    A new Columbia and MIT paper reports that Claude autonomously discovered a faster method for computing selected entries of a product of two long, thin matrices, which classic reductions turn into speedups for 3SUM and all-pairs shortest paths. 3SUM now runs in O(n^1.9992) and APSP in O(n^2.9995, tiny gains that nonetheless refute the long-standing 3SUM and APSP hypotheses. The method was found in a single AI session using 16M output tokens with no human input and was formally checked in Lean 4.

  • Hacker News AIT2·Open Source·71 pts

    The open source LookAtMyQuant project introduces a post-training quantization compiler that allocates bit budgets per stored unit based on measured output damage, using Fisher information and trellis codes. Its second version compresses Qwen3.5-4B to 1.23 GB (~2.3 bits per weight), solving 330 of 540 exact-answer math problems versus 328 for the same-size Bonsai 2 Ternary 4B, with lower KL divergence (0.45 vs 0.59 nats). The full recipe and roughly $34 of GPU cost are published under the MIT license.

  • Arena.ai (@arena)T1·Models·58 pts

    OpenAI's GPT-6.2 Astra Gains Coding Strength but Loses Chat Ground

    Original: GPT-6.2 Astra (Max) by @OpenAI ranks #6 on the Agent Arena Index, one place above GPT-6 Astra (Max).

    Arena.ai's Agent Arena Index places OpenAI's GPT-6.2 Astra (Max) at #6 overall, just one spot above GPT-6 Astra (Max). The new model improves coding from 1,288 to 1,338, rising from #8 to #3, but drops in chat from 1,265 to 1,211, falling from #8 to #11, with the two shifts nearly canceling out in the aggregate score.

  • Hacker News AIT2·Tools·58 pts

    Open-source project gopgext uses a custom TinyGo target and runtime layer to compile plain Go code into PostgreSQL extensions, auto-generating C glue, install scripts and control files. It supports the standard library, SQL type mapping, SPI queries and SQLSTATE error codes, but remains experimental and is explicitly not production-ready.

  • Rohan Paul (@rohanpaul_ai)T2·Models·88 pts

    Anthropic mathematician Alpöge used Claude to prove the Riemann Hypothesis, formalizing the entire argument into roughly 17 million lines of machine-checkable code that Lean's official community certified within a single day. It marks the first time AI has helped crack mathematics' most famous open problem after 167 years, with Claude proposing a key step in areas beyond the human's own expertise.

  • LangChain BlogT1·Agents·58 pts

    LangChain Pitches Knowledge Factory Architecture for Enterprise Agents

    Original: How to build a knowledge factory for AI agents

    LangChain introduces the "knowledge factory" concept, arguing that the bottleneck for enterprise AI is not model capability but capturing, curating and maintaining the tacit operational knowledge scattered across employees and documents into a representation agents can consume. The architecture is already used internally across LangChain's GTM and content teams and is being rolled out to select customers.

  • Rohan Paul (@rohanpaul_ai)T2·Models·78 pts

    NVIDIA Method Caches LLM Reasoning as CPU Code to Cut Costs 291x

    Original: New NVIDIA paper finds that once an LLM has reasoned through a problem, the solution can run as CPU ...

    An NVIDIA paper proposes compiling an LLM's reasoning trace into small parameterized Python functions that run on CPU, invoking a fine-tuned 1B model only for lightweight text tasks such as value extraction. The cache matched or beat the source model on 7 of 8 math and science benchmarks, scoring 93.3% versus 83.3% on AIME '25, and covered at least 90% of rephrased or re-numbered variants at 16x to 291x lower cost.

  • The DecoderT2·Models·58 pts

    Nvidia Researchers Call Huge Context Windows a Dead End

    Original: Nvidia researchers argue huge AI context windows are a dead end

    An Nvidia research team argues that ever-larger context windows are often overkill and expensive. In many tests, periodic summarization combined with targeted lookups performed about as well as stuffing entire documents into the prompt, suggesting the industry should rethink the 'longer is better' assumption.