AI News
AI community posts, news and release updates
Latest updates
- Rant
Asked Gemini for hiking gear, got a survival kit for when I get lost – seriously?
Published Jul 23, 2026A user asked Google's Gemini AI what to bring for hiking, and the AI assumed they would get lost, providing a list of emergency supplies. The user felt disrespected and complained about the response.
- Gemini
- 登山
- AIの限界
- ユーザー体験
- News
AI Daily Recap: White House Accuses Moonshot of Distilling Fable; OpenAI Plans $750B Spending; AMD Invests in Anthropic
Published Jul 23, 2026Daily AI news roundup: White House OSTP director claims Moonshot's Kimi K3 distilled from Claude Fable and violated export controls; nearly 200 Silicon Valley companies sign open letter against blocking Chinese models; OpenAI projects $750B compute spending through 2030; AMD and Anthropic make compute-for-investment deal; OpenAI launches Presence platform; SpaceXAI building Texas data center; Samsung invests in Mistral; Travis Kalanick launches Atoms robotics; Trump nominates tech critic for DOJ antitrust; Stripe revenue soars; Google raises capex forecast; Gemini reaches 950M MAU.
- Anthropic
- OpenAI
- Distillation
- policy
- data center
- Moonshot AI
- AMD
- White House
- Idea
Wenfeng's belief that AGI labs won't capture most value is a coherent strategic position aligned with Chinese AI ecosystem
Published Jul 23, 2026The author argues that Wenfeng's belief that AGI labs won't capture most value is not idealism but a coherent strategic position aligned with Chinese AI ecosystem characteristics: intelligence is just another industrial input.
- AGI
- DeepSeek
- 价值捕获
- Idea
Loops are a short-lived patch; new models can do long-horizon reasoning out of the box
Published Jul 23, 2026The author argues that loops were a short-term fix for models that couldn't reliably work on long problems until a goal; new models like Fable and GPT-5.6 can do it out of the box.
- GPT-5.6
- Fable
- Model Capability
- Kimi K3
- AI loops
The author infers from reactions by Trump, the White House, Jensen Huang, and others that Kimi K3's lead has received a "collective certification of US losing composure," claiming this is even more intense than the DeepSeek moment.
- Moonshot
- AI竞争
- Kimi K3
Merge Gateway offers a free plan with $10 monthly credits ($20 first month), access to 200+ models including Claude Fable 5 and GPT-5.6, with a 5% markup on provider cost. Phone verification required.
- 免费API
- Merge Gateway
- 模型聚合
LitigationBench is a new benchmark for evaluating AI performance on litigation tasks.
- AI benchmark
- LitigationBench
- News
Claude Opus 5 tomorrow? Performance on par with Fable 5, price and availability pending.
Published Jul 23, 2026Rumors suggest Anthropic will release Claude Opus 5 tomorrow, with performance comparable to Fable 5. Pricing and availability details are still unconfirmed, generating significant interest.
- Anthropic
- Fable 5
- AI model release
- Claude Opus 5
The author criticizes US double standards: Chinese labs learning from US models is theft, but an ex-OpenAI CTO using DeepSeek architecture and data is "engineering." China opens sources, lowers costs; US blocks, overcharges, then cries IP theft.
- 开源
- 蒸馏
- AI竞争
- Idea
Liang Wenfeng's 4-hour talk: AGI roadmap, continual learning, restraint as strategy
Published Jul 23, 2026In an investor meeting, Liang Wenfeng detailed DeepSeek's AGI roadmap: CoT → Agent → continual learning → AI self-iteration → embodied intelligence. He emphasized fair profit, open source as concession, cost efficiency, and predicted consolidation of China's fragmented AI resources.
- 开源
- AGI
- DeepSeek
- 梁文峰