AI Daily Industry Briefing | 2026-08-18

Jeacson She8/18/26...About 5 min

AI Daily Briefing

2026-08-18 | 🇨🇳 中文 + 🇺🇸 English Past 24h focus: AI agents & open-source LLMs

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

harry0703/MoneyPrinterTurbo ⭐105,987 — Python | One-click AI short-video generation — Built on AI large models and automated workflows, it generates HD short videos from just a topic or keyword
santifer/career-ops ⭐64,631 — JavaScript | Open-source AI job-hunting assistant — Scans job platforms, scores them on an A–F structure, tailors your résumé, and tracks application progress
usestrix/strix ⭐54,157 — Python | Open-source AI penetration testing — An AI penetration-testing tool that automatically finds and fixes application security vulnerabilities
AlexsJones/llmfit ⭐32,259 — Rust | Find a runnable model with one command — Covers hundreds of models and providers to help you find models that can run on your local hardware
mukul975/Anthropic-Cybersecurity-Skills ⭐28,411 — Python | Cybersecurity skill library for AI agents — 817 structured cybersecurity skills, mapped to six major frameworks including MITRE ATT&CK and NIST CSF
jundot/omlx ⭐18,986 — Python | Apple Silicon LLM inference service — A local LLM inference server with continuous batching + SSD caching, managed from the macOS menu bar
akitaonrails/ai-memory ⭐2,042 — Rust | Long-term memory for agent CLIs — Provides long-term memory for agent coding CLIs and supports task handoff between different agent vendors

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

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

Twin: Playing an Unknown Game with a Test-Time Digital Twin | cs.AI | Alexy Skoutnev, Kirill Acharya Lets frontier coding agents write an executable "digital twin" world model at test time to play an unknown game — a new paradigm of agents building models autonomously (submitted 08/14)
Handover of In-Context Learning State Across Session Boundaries | cs.AI | Masahiro Kato, Taka Kato Studies methods and theoretical properties for handing over in-context learning state across session boundaries in LLM applications, striking directly at the pain point of long-horizon agent tasks (submitted 08/14)
Split the Labor: Separating Evidence Interpretation from Decision Aggregation | cs.AI | Zhelun Wu Separates "evidence interpretation" from "decision aggregation" to improve the quality of conclusions from multi-source LLM reasoning (submitted 08/14)
You Only Pass Once: Answering and Abstaining Together in a Single Forward Pass of a Frozen Language Model | cs.CL | Ziyang Luo, Zhongyao Chu A frozen LLM answers and abstains simultaneously in a single forward pass, easing the underuse of evidence in reasoning tasks (submitted 08/14)
Approximate Muon with low-rank adapters | cs.LG | Ben Anson, Conor Houghton Approximates the Muon optimizer with low-rank adapters, reducing memory overhead in the pre-training/fine-tuning stage (submitted 08/14)

Note: the latest submission date is 08/14 (fewer submissions over the weekend); the above are the 5 papers most relevant to agent / reasoning / training.

🐦 Twitter/X — Tracked Accounts

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

🔴 Key Signals (within 24h)

@NousResearch (08/18 03:09) — Releases Hermes Desktop Bot Mode: agent profiles become named Bots, each with its own role, model, memory, skills, and avatar
@Teknium (08/18 03:14) — Introduces Hermes Agent Bot Mode again: replacing sessions with "bots," one chat per agent profile, letting you assign tasks to a bot
@openclaw (08/18 06:30) — Retweets Pat_Erichsen: x402 🤝 OpenClaw partners with AWS to launch the AWS Agents Pay plugin, supporting agentic payments
@dotey (08/18 02:01) — Cursor's code-hosting platform Origin officially launches: it can sync repos from GitHub, an "AI-era GitHub" designed around the cadence of agent collaboration
@dotey (08/18 01:32) — yetone open-sources Cumora: it turns AI Agents into full members of chat groups — able to DM, create groups, claim tasks, and send/receive real emails
@dotey (08/18 08:11) — Retweets Phoenix Tech: a Meituan executive reviews the company-wide "shrimp-farming" AI overhaul — bills spiked, burning tens of millions of yuan a day, and the errors even disrupted real operations
@OpenAI (08/17 20:22) — gdb: defenders can see the future, the window is narrow, and you must upgrade your cybersecurity fundamentals with the best AI tools now

Notable Posts (older than 24h but strong signal this week)

@OpenAI (08/14 01:01) — Previews Ultrafast mode: GPT-5.6 Sol up to 14x faster, opening first to select API customers
@GoogleAI (08/14 01:06) — Releases Gemini 3.7 Flash: the strongest "workhorse" model to date for coding and agents, already rolled out on the Gemini App / Enterprise Agent Platform
@AnthropicAI (08/15 03:16) — Publishes a watermarking FAQ: implementing content watermarking for EU AI Act compliance, answering community questions about watermarking
@OpenAI (08/14 04:15) — ChatGPT desktop adds Computer History: remembers your action history across apps/websites, more personalized with lower token use
@Kimi_Moonshot (08/11 16:20) — Kimi K3 (open-weight, strong at agentic coding) launches on Databricks, after previously entering GitHub Copilot

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

@tonbistudio (08/17 13:41) — Watching a team of bots automatically split up a large game project purely by their own experience and expertise — "a bit crazy, and a bit scary"
@Stephan_Talk (08/18 00:42) — On AI software engineering: AI can't know "the things you should have told it but didn't"; requirements and architecture up front remain essential
@imbabybrooklyn (08/18 04:44) — Loves Hermes Agent's curated gen UI: for skills like /get-price btc it renders a live inline widget
@joshwoodward (08/15 03:19) — Gemini 3.7 Flash has landed in the Gemini App
@genevieve__h (08/14 22:52) — Gemini 3.7 Flash powers Spark: higher tool-call success rates, an underrated work assistant

📰 Weibo Highlights

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

@星熠落云舟 (08/18 08:21) — #大批AI博主停更了#: in August AI drama and AI tutorial bloggers across the web announced a wave of shutdowns, as the traffic era of template-based mass account-farming and crude arbitrage comes to an end
@亢岳 (08/17 19:51) — Feels "physiologically uncomfortable" seeing AI faces: similar in form but not in spirit, like pre-made meals — edible but not up to par
@雪珛 (08/18 08:14) — America "goes all in": tech giants keep piling into compute and large models, and US equities are highly bound to AI expectations
@宋嘉红观势 (08/18 08:13) — Embodied-AI financing heats up: Beijing's CIFTIS sets up a "Robots+" exhibition zone as primary-market capital pours in
@karminski-牙医 (08/16 11:49) — DeepSeek-V4-Pro-0813 hands-on: burned 100M tokens verifying where problems like "max reasoning effort even worse than high" come from
@karminski-牙医 (08/14 20:07) — Questions GLM-5.3's test scores "6x'ing": Terminal Bench 3.0's task difficulty is the real deal
@karminski-牙医 (08/14 19:12) — "Qwen3.8-27B is about to ship too… can't keep up"

🌐 Blog Picks

(Past 36h, unread, AI/LLM topics)

How canvases make agentic workflows visible, steerable, and cost-efficient | The GitHub Blog | 08/17 — GitHub Copilot uses canvases to make agentic workflows visible, steerable, and more cost-efficient
The Defender's Window | OpenAI News | 08/17 — An OpenAI cybersecurity long-read: the attack-defense window is narrowing, and defenders must upgrade their fundamentals now with the best AI tools (same source as the gdb tweet)
OpenAI joins PORTS-Pike project | OpenAI News | 08/17 — OpenAI joins the PORTS-Pike port-security AI project
New policy ideas for the Intelligence Age | OpenAI News | 08/17 — OpenAI proposes new policy ideas for the "Intelligence Age"
Stripe's AI Strategy: Build More, Not Less | The a16z Show | 08/17 — a16z podcast: Stripe's AI strategy is "build more, not save more"

🎯 One-Line Summary of the Day

"Agent infrastructure bloomed overnight: Hermes launched Bot Mode, Cursor rolled out Origin code hosting, OpenClaw connected agentic payments, and Gemini 3.7 Flash rolled out broadly; meanwhile the wave of Chinese AI bloggers going dark and Meituan's 'shrimp-farming' reflections are drawing a line under the AI arbitrage era."

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