2026年7月22日
AI 消息
这周值得知道的 AI 消息。
Will AI help you do your job or replace you?
BBC charts analyze how AI is impacting jobs, exploring claims made by AI companies about replacing human labor.
看原文 →Headaches for Silicon Valley as China chips away at the US’s lead in the AI race
China's advancements in AI are challenging US tech dominance, raising concerns in Silicon Valley.
看原文 →Meta is testing an AI bedtime story app for people with no imagination
Meta is testing a new AI-powered app called StoryKit that generates personalized bedtime stories for children. The app is currently limited to specific regions as Meta gathers feedback from parents.
看原文 →An AI system helped Pakistani judges clear massive backlogs at $38.50 return per dollar invested
A field experiment in Pakistan found that an AI assistant called JudgeGPT boosted case resolution by 6.3% for judges who received training, leading to a high return on investment.
看原文 →Google ships Gemini 3.6 Flash, 3.5 Flash-Lite and 3.5 Flash Cyber
Three new Gemini models built for agent work, designed to run cheaper and spend fewer tokens.
看原文 →Anthropic’s $1.5B copyright settlement approved; only 350 authors opted out
A US judge approved Anthropic's $1.5 billion copyright settlement, with only 350 authors opting out despite an initial attempt to block opt-outs.
看原文 →Introducing the ChatGPT for small business program
OpenAI launched a program called ChatGPT for Small Businesses to help entrepreneurs use ChatGPT Work for automation and growth.
看原文 →配音与开源
现在值得一看的开源语音、配音工具。
dramaclaw/dramaclaw
dramaclaw is a new AI tool on GitHub that generates videos from scripts across various genres including dramas, ads, and even otome games.
看原文 →
Kritt-ai/open-kritt
Kritt-ai/open-kritt, a new open-source project on GitHub, allows developers to use AI agents for identifying security vulnerabilities in code.
看原文 →本周 X 上的讨论
这周 AI 圈在 X 上聊得最多的。
Sam Altman (@sama)
There was a serious security incident while our models were being evaluated. Sharing what we know so far. Thanks to @huggingface for working through it with us.
看原文 →Arena.ai (@arena)
Good news: Kimi K3 lands at #4 on the Agent Arena leaderboard, shoulder to shoulder with Claude Opus 4.8 and GPT-5.6 Sol. Once the weights drop on July 27 as planned, it becomes the top open-weights model.
看原文 →Jun Song (@jun_song)
Where open-source AI stands: GLM-5.2 is nearly Opus level. Kimi-K3 is nearly Fable level. Qwen-3.8 at 2.4T params should pass Opus. Deepseek V4 shipped at $0.0028 per million tokens and should pass Opus too. Minimax-M3-Pro at 3T params looks Fable level. GLM-5.5 is expected to pass Opus.
看原文 →Bilawal Sidhu (@bilawalsidhu)
Translation: the 20x plan is now the 10x plan, and the 5x plan is the 2.5x plan. Congrats on your upgrade. (quoting @claudeai: "Beginning July 20, Claude Fable 5 will be included in all Max and Team Premium plans, at 50% of limits.")
看原文 →用邮件接收下一期
8 种语言任选，随时可以退订。