🤖 AI 资讯日报 — 2026-06-21

自动生成于 2026-06-21 08:32 UTC+8 | 数据源: T1 官方 + T1.5 媒体 + T2 社区

🔴 T1 官方一手

最高权重，直接来自官方渠道

OpenAI Blog (3 条)

• Codex for every role, tool, and workflow
• Discover new Codex plugins, sites, and annotations that help analysts, marketers, designers, investors, and other teams
• Introducing GPT-5.5
• Introducing GPT-5.5, our smartest model yet—faster, more capable, and built for complex tasks like coding, research, and
• Sora 2 is here
• Our latest video generation model is more physically accurate, realistic, and controllable than prior systems.

Anthropic Research (5 条)

• Project Fetch: Phase two
• We report results from our latest test of whether Claude can help Anthropic employees perform sophisticated robotics tas
• Claude Fable 5 and Claude Mythos 5
• Today we're launching Claude Fable 5: a Mythos-class model that we've made safe for general use.
• When AI builds itself
• For most of AI's history, humans drove every step in its development cycle. But at Anthropic, we are delegating a growin
• Agentic coding and persistent returns to expertise
• Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems
• Introducing Claude Opus 4.8
• Our latest model, Claude Opus 4.8, is an upgrade to our Opus class of models, with stronger performance across coding, a

HuggingFace Blog (5 条)

• Beyond LoRA: Can you beat the most popular fine-tuning technique?
• We're on a journey to advance and democratize artificial intelligence through open source and open science.
• Introducing North Mini Code: Cohere's First Model For Developers
• A Blog post by Cohere Labs on Hugging Face.
• GLM-5.2: Built for Long-Horizon Tasks
• A Blog post by Z.ai on Hugging Face.
• Is it agentic enough? Benchmarking open models on your own tooling
• We're on a journey to advance and democratize artificial intelligence through open source and open science.
• Agentic Resource Discovery: Let agents search for tools, skills, and other agents.
• We're on a journey to advance and democratize artificial intelligence through open source and open science.

GitHub Blog (4 条)

• How we built an internal data analytics agent
• Learn how GitHub built Qubot, our internal Copilot-powered analytics agent, to allow any GitHub employee to ask question
• AI credits consumed per user now in the Copilot usage metrics API
• The Copilot usage metrics API now reports how many AI credits each user consumed per day, derived from the same AI credi
• MAI-Code-1-Flash available on more Copilot surfaces
• MAI‑Code‑1‑Flash, Microsoft's purpose‑built small coding model, is now available across additional GitHub Copilot surfac
• Getting more from each token: How Copilot improves context handling and model routing
• How GitHub Copilot is making more of each session go toward useful work, so your credits go further.

Apple ML (1 条)

• The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity
• Recent generations of frontier language models have introduced Large Reasoning Models (LRMs) that generate detailed thin

NVIDIA Blog (2 条)

• At Cannes Lions, NVIDIA Partners Reshape Advertising and Marketing With AI
• The digital era gave the advertising and marketing industry speed; the AI era is giving it autonomous operations.
• France Advances Europe's AI Future With NVIDIA Technologies
• A year ago at NVIDIA GTC Paris at VivaTech, France laid out plans to advance local AI — from new AI factories and nation

xAI (1 条)

• Grok Imagine Video 1.5
• Grok Imagine Video 1.5 is now generally available on the Imagine API. Video 1.5 Fast almost doubles generation speed: it

Simon Willison (2 条)

• GLM-5.2 is probably the most powerful text-only open weights LLM
• Chinese AI lab Z.ai released GLM-5.2 to their coding plan subscribers on June 13th, and then publicly on June 16th. Simo
• A quote from Sean Lynch on MCP authentication isolation
• The real valuable capability MCP offers over skills/CLI is isolating the auth flow outside of the agent's context window

🟡 T1.5 媒体

主流科技媒体和开发者社区

THE DECODER (2 条)

• Microsoft researcher builds a working neural network out of goats in Age of Empires II to critique AI science
• Adrian de Wynter, a researcher at Microsoft and the University of York, has built a working neural network inside the ma
• ChatGPT's new health upgrade beats doctor-written answers, OpenAI says
• OpenAI has upgraded ChatGPT's healthcare capabilities with GPT-5.5 Instant. In the company's own comparative tests, the

ITHOME (3 条)

• 2026世界人工智能大会今年7月在上海举办，超300款AI产品全球首发
• 2026年世界人工智能大会将于7月17日至20日在上海举办，主题为「智能伙伴，共创未来」，超300款AI产品将全球首发，由外交部、科技部、工信部等10个部门与上海市政府共同主办。
• 上交所：AI大模型企业适用科创板第五套上市标准
• 上海证券交易所发布指引，支持尚未形成一定收入规模的优质人工智能大模型企业在科创板发行上市，要求至少有一个大模型产品已完成上线发布并实现规模化应用。
• Anthropic高管与美政府展开谈判，寻求解除AI模型Fable 5出口禁令
• Anthropic公司高层与特朗普政府官员就Fable 5模型出口禁令展开面对面谈判，详细介绍了其网络安全防护机制，希望推动限制解除。

HackerNews (15 条)

• Ask HN: Is it time to fork HN into AI/LLM and "Everything else/other?" 👍553
• Points: 553, Comments: 372
• Coconut by Meta AI – Better LLM Reasoning with Chain of Continuous Thought? 👍362
• Points: 362, Comments: 156
• Show HN: Countless.dev – A website to compare every AI model: LLMs, TTSs, STTs 👍361
• Points: 361, Comments: 76
• Taste in the age of AI and LLMs 👍265
• Points: 265, Comments: 211
• Wikimedia Enterprise – APIs for LLMs, AI Training, and More 👍222
• Points: 222, Comments: 161
• Anti-AI Hype LLM Reading List 👍208
• Points: 208, Comments: 52
• Jellyfin LLM/"AI" Development Policy 👍207
• Points: 207, Comments: 108
• Ask HN: Go deep into AI/LLMs or just use them as tools? 👍195
• Points: 195, Comments: 133

🟢 T2 社区

KOL 观点、学术论文、社区讨论

KOL Twitter (7 条)

• LLM Cheat-Sheet for OpenClaw and Hermes agents
• The LLM Cheat-Sheet for OpenClaw and Hermes agents. 12 models, 4 tiers. Tier 1: Claude Opus 4.6, GPT-5.4, Gemini 3.1 Pro
• Complete Guide to Building Crypto AI Agents in 2026
• The reasoning layer is the LLM, whether that is Claude, GPT, Gemini, or an open source model.
• How to Build a Moat in AI Agent Trading
• AI is building trading agents. GPT-4, Claude, Gemini, open-source fine-tunes.
• Deep Agents: structuring planning for multi-step problems
• Most agents today are shallow. We're entering the era of Deep Agents - systems that strategically plan, remember, and de
• How to finally trust Claude's advice (using Karpathy's LLM Council)
• His version polls GPT, Claude, Gemini, and others against each other. I rebuilt it to work entirely inside Claude using
• Andrew Ng on agentic workflows and LLM function calling
• As agentic workflows mature, many developers are prompting LLMs to carry out agentic behaviors. Even before GPT-4 native
• Jiaxuan You: LLM agents manage a portfolio
• LLM agents manage a portfolio of stocks, Polymarket positions, and cash. Track how state-of-the-art LLMs (GPT, Claude, G

ArXiv (10 条)

• How Transparent is DiffusionGemma?
• LLM reasoning transparency is a critical affordance for understanding model decisions, mitigating misuse and misalignmen
• UNIEGO: Proxies as Mediators for Unified Egocentric Video Representation Learning
• Egocentric video understanding is inherently limited by the narrow perspective of wearable cameras: a single viewpoint,
• Optimal Deterministic Multicalibration and Omniprediction
• A model is multicalibrated on a collection of group weights $G$ if it is calibrated -- i.e. unbiased even conditional on
• Structuring and Tokenizing Distributed User Interest Context for Generative Recommendation
• Generative recommendation is an emerging paradigm that has shown promise in industrial recommendation systems, aiming to
• The Token Is a Group Element: On Lie-Algebra Attention over Matrix Lie Groups
• We place the attention token on the group: a token is an element $g_i$ of a matrix Lie group $G$ -- a bare transformatio
• Predictability as a Fine-Grained Measure for Privacy
• Differential privacy (DP) ensures rigorous individual-level privacy guarantees against even the most knowledgeable attac
• Toward Calibrated Mixture-of-Experts Under Distribution Shift
• Calibration aligns a model's predictive uncertainty with the frequencies of its empirical outcomes and is important for
• Multi-Task Bayesian In-Context Learning
• Bayesian predictive inference provides a principled framework for uncertainty quantification, data efficiency, and robus

HF Daily Papers (15 条)

• LedgerAgent: Structured State for Policy-Adherent Tool-Calling Agents ⭐6
• Policy-adherent tool-calling agents in customer-service domains must maintain task states across turns while calling too
• Context-Aware RL for Agentic and Multimodal LLMs ⭐9
• Large language models (LLMs) often fail when answering requires identifying a small but decisive piece of evidence withi
• Freeing the Law with LOCUS: A Local Ordinance Corpus for the United States ⭐6
• Progress in legal AI increasingly depends on access to authoritative legal text at scale. Yet one of the most consequent
• ReSyn: A Generalized Recursive Regular Expression Synthesis Framework ⭐2
• Existing Programming-By-Example (PBE) systems often rely on simplified benchmarks that fail to capture the high structur
• LegalHalluLens: Typed Hallucination Auditing and Calibrated Multi-Agent Debate for Trustworthy Legal AI ⭐4
• AI systems deployed in legal workflows hallucinate at rates that aggregate metrics report at ~52%, but this average conc
• Configurable Clinical Information Extraction with Agentic RAG: What Works, What Breaks, and Why ⭐4
• Patient contexts span hundreds of heterogeneous documents and thousands of structured data points, yet the document-leve
• The Data Manifold under the Microscope ⭐3
• A significant gap exists between theory and practice in deep learning. Generalization and approximation error bounds are
• The FID Lottery: Quantifying Hidden Randomness in Generative-Model Evaluation ⭐4
• The Frechet Inception Distance (FID) is the de facto arbiter of image generation, yet most papers report just a single n

本日报由 Hermes Agent 自动生成，数据截至 2026-06-21。如有遗漏，欢迎反馈。