AI Daily Industry Briefing | 2026-11-14

Jeacson She11/14/26...About 6 min

AI Daily Briefing

2026-11-14 | 🇨🇳 中文 + 🇺🇸 English Past 24h focus: AI agents & open-source LLMs

(Today's AI-related trending repos, the 8 most relevant to agents / LLM / training and inference infrastructure by priority)

alibaba/open-code-review — Go | ★215 today | Hybrid code review: deterministic pipelines + LLM agent — Alibaba's open-source code review tool: a hybrid architecture of deterministic pipelines plus an LLM agent, with line-precise comments and built-in multi-language rule sets for NPE / thread safety / XSS / SQL injection, compatible with both the OpenAI and Anthropic APIs.
GoogleChrome/modern-web-guidance-src — TypeScript | ★1,925 today | Keep your coding agent up to date with web best practices — A rule source that feeds coding agents the latest web best practices; one of the biggest star gainers of the day.
openai/codex-security — TypeScript | ★751 today | Codex Security CLI + SDK for finding and fixing vulns — OpenAI's official security CLI and TypeScript SDK: find, validate and fix security vulnerabilities, pluggable straight into agent pipelines.
ZhuLinsen/daily_stock_analysis — Python | ★256 today | LLM-powered multi-market stock analysis system — An LLM-driven multi-market stock analysis system: multi-source quotes, real-time news, a decision dashboard and automatic push, pitched at zero-cost scheduled runs.
backnotprop/plannotator — TypeScript | ★108 today | Workspace to shape, annotate and review agent plans — A workbench that keeps humans in the agent loop: plan, annotate, review code, and send feedback back to the agent in one click.
tradermonty/claude-trading-skills — Python | ★69 today | Claude Code skills for equity investors — A Claude Code skills collection for equity investors: market analysis, technical charts, economic calendar, stock screening and strategy development.
Likely7/Veyra-NRVideo — C++ | ★256 today | AI super-resolution + frame generation video player — A Windows video playback and capture enhancement tool layering AI super-resolution, DLSS 5 image enhancement and frame generation; the name is heavy on marketing, but it is a genuine sample of on-device inference shipping for real.
ChromeDevTools/chrome-devtools-mcp — TypeScript | ★67 today | Chrome DevTools for coding agents — The official packaging of Chrome DevTools as an MCP, letting coding agents read pages and run debugging directly.

📄 arXiv Papers — cs.AI / cs.LG / cs.CL

(The 5 papers most relevant to agents / reasoning / LLM training / inference)

AgentSSL: Can MLE Agents Leverage Unlabeled Data? | cs.AI | Ethan Lin, Akanksha Sarkar Asks whether machine-learning engineering (MLE) agents can use unlabeled data the way human engineers do, easing the two bottlenecks in new domains: acquiring labels and hand-building pipelines.
CoReMem: Cost-Internalizing Memory for Successive Coding Tasks with Prospective Valuation | cs.AI | Peiwen Li, Harshit Verma Coding agents work successively on an evolving codebase while the next task is unknown; the paper turns "what memory to write" into a valuation problem that internalizes future cost.
Classical Energies Learn to Reason 1,000x Faster than Looped Models | cs.AI | Benjamin Hoover, Hugh Blayney Replaces looped structures with classical energy models for reasoning, claiming faster and more stable training with speed-ups of up to three orders of magnitude.
Negative-Only Policy Optimization for One-Sided Verifiable Rewards | cs.LG | Jiacheng Xu, Shuo He Adapts RLVR to the realistic case where a verifier can only judge wrong, not right: the policy can still be optimized from negative-sample signals alone.
Metal Sandbox: Isolated Execution and Command Scheduling for GPU Coding Agents on Apple Silicon | cs.LG | Haozhe Wang, Lina Chen Provides isolated execution and command-scheduling sandboxes for GPU coding agents on Apple Silicon, addressing the automated training and evaluation of on-device inference kernels.

(Note: the latest submission date in this arXiv batch is 11/12, not today's new submissions.)

🐦 Twitter/X — Tracked Accounts

(From 10 tracked accounts: @AnthropicAI @OpenAI @GoogleAI @GeminiApp @sama @Kimi_Moonshot @NousResearch @openclaw @Teknium @dotey)

🔴 Key signals (within 24h)

@AnthropicAI (11/14 03:00) — The engineering blog publishes 10 recommendations for "designing and running safety evaluations", spelling out when a model may be deployed and which guardrails it should carry · 原文
@AnthropicAI (11/14 00:00) — JPMorgan CEO Jamie Dimon and Dario Amodei in a livestreamed fireside chat (IPO roadshow context; the tweet carries an SEC prospectus disclaimer) · 原文
@GeminiApp (11/14 06:05) — With a one-line brief plus a custom skill, Gemini 4 Hydrogen finds public-domain footage on its own and cuts it into a complete narrated short film · 原文
@Google (11/14 01:27) — Gemini personal intelligence plugs into Google Wallet: generate cash-flow and asset-allocation charts in conversation, find subscriptions you forgot you had, set a holiday budget from last year's data · 原文
@witcheer (retweeted by @NousResearch) (11/13 19:29) — An IT director tops up Nous Portal with $10 and lets Hermes Agent research on its own, download ISOs and stand up an entire Windows domain (2 DCs + print/file/app servers); the actual spend was 27 cents · 原文
@modal (11/14 01:04) — Modal Endpoints plugs into Hermes Agent as a plugin: hermes plugins install modal · 原文
@openclaw (11/13 12:31) — Gemini 4 Hydrogen added to OpenClaw on launch day · 原文

Notable older posts (outside 24h but strong signal this week)

@GoogleAI (11/13 02:03) — Gemini 4 Hydrogen formally released: its default thinking tier sits above Argon, it runs 2.5× faster than 3.8 Flash, and it can be paired with Carbon inside Antigravity · 原文
@jeffreygwang (retweeted by @OpenAI) (11/12 02:17) — New OpenAI research: as models get stronger, how CoT monitorability (non-CoT capability and controllability) evolves; the conclusion is "unsurprising but sobering" · 原文
@AnthropicAI (11/12 19:30) — Formally kicks off its IPO (registration statement filed with the SEC) · 原文
@openclaw (11/12 00:12) — Claude Fable 5.2 added to OpenClaw; Anthropic says output is up to 30% faster than Fable 5.1 · 原文
@cerebras (retweeted by @Kimi_Moonshot) (11/07 03:14) — Runs a full online shopping flow with Kimi K3.1 Pro: one prompt → find the product → checkout → order confirmation, in 11.5 seconds · 原文

🐦 Twitter/X — Popular Discussions (broad search)

(agent / open-source LLM topics, top 5)

@ArtificialAnlys (11/13 02:07) — Gemini 4 Hydrogen scores 54 on the intelligence index, one point behind Claude Sonnet 5.5 and GPT-6.1 Sol, at roughly 32% of Sonnet's cost; the new xhigh tier is the highlight · 原文
@googledevs (11/14 06:08) — Antigravity ships Orchestrator mode: Carbon/Argon read dependencies and produce a plan, Hydrogen writes code and runs tests fast, and when a test fails the frontier model comes back to revise the plan · 原文
@sundarpichai (11/14 04:23) — This week's shipping list: two eighth-generation TPUs, Gemini 4 Carbon, AlphaProof Code, Gemini Robotics 2.5, AlphaGenome 2, Gemini Decisions API, Gemini 4 Hydrogen · 原文
@studio_yebisu (11/13 03:06) — A Japanese developer's hands-on: Hydrogen is extremely fast, genuinely multimodal, and natural in Japanese; already considering a mixed stack of "Fable 5.2 designs + Opus 5.5 directs + Hydrogen does the work" · 原文
@0xrux (11/13 16:41) — Open-source AI gateway FreeLLMAPI: 34 free providers, 635 free endpoints, roughly 7.4 billion tokens per month, with built-in rate-limit failover and coding-agent support · 原文

📰 Weibo Highlights

(High signal-to-noise channels: karminski-牙医, 宝玉 and other AI bloggers)

@IT之家 (11/14 08:14) — llama.cpp and the ggml libraries reach version 1.0.0: support for over 150 model architectures, running at full strength locally even offline · 原文
@爱可可-爱生活 (11/14 08:12) — Breaks down Antigravity's "one entrance, two brains": static routing picks a model when the task starts, while Orchestrator swaps models dynamically by stage — reading code, mapping dependencies, settling the plan · 原文
@暗流情报 (11/14 07:41) — NVIDIA's board announced on 11/12 that Jensen Huang is stepping down as president, CEO and director, with CFO Colette Kress as interim CEO; the stock fell as much as 13% · 原文
@karminski-牙医 (11/12 09:19) — MiniMax-H3-Image hands-on: 4-megapixel output in 140 seconds on an M2 Ultra, versus 15 minutes for QwenImage2.1; detail still lags between 6 step and 20 step, better suited to vector-style art · 原文
@karminski-牙医 (11/10 12:11) — Fable-5.2 versus deepseek-v4.1-flash on the same task: one spends $40, the other ¥1.4 · 原文
@karminski-牙医 (11/10 08:01) — Tencent Youtu releases the draft model Qwen3.8-27B-DSpine-b16, lifting speculative decoding from 1.6× with MTP to 3×, at 805 tps (8 concurrent) · 原文

🌐 Blog Picks

(Published in the past 36h, unread, AI/LLM topics)

ChatGPT Wallet lets OpenAI's Dots agents make purchases for you | The Verge | 11/13 — ChatGPT Wallet lands: OpenAI's Dots agents can place orders on the user's behalf, making the boundary of payment authority the focus of discussion.
Introducing ChatGPT Wallet | OpenAI News | 11/13 — The official product announcement; the account and authorization design for agent-mediated payments is defined here.
Apple Releases tvOS 27.2.1 With Siri AI for Older Apple TVs | MacRumors | 11/13 — Siri AI comes down to older Apple TVs, extending the reach of on-device assistants.
Siri AI Still Not Enabled on HomePod Mini 2 With Latest Software Update | MacRumors | 11/13 — The new HomePod mini 2 still doesn't get Siri AI: the hardware was updated, the model wasn't.
Hands-On With the New Apple TV 4K and Siri AI | MacRumors | 11/12 — Hands-on: how usable Siri AI really is in the living room, and where it falls short.
研究生如何用 AI 辅助完成一篇综述：我的科研写作工具与实践经验 | 少数派 | 11/13 — A complete workflow from search and screening to final draft, useful as a sample of agent-assisted academic writing.
在快 AI 时代，我是怎么把 Mole 做成一款「慢产品」的？ | 少数派 | 11/13 — Thinking in the opposite direction: once AI speeds everything up, why a product experience should deliberately slow down.

🎯 One-line Takeaway

"Gemini 4 Hydrogen turns 'cheap worker model + frontier model as planner' into the default architecture; the cost structure of agents is being repriced."

Generated automatically by Hermes Agent | Sources: GitHub Trending · arXiv · Twitter/X · Weibo · RSS · HF Papers