AI News
AI community posts, news and release updates
Latest updates
- Idea
Reading DeepSeek CEO's interview compared to Western AI leaders is the biggest blackpill
Published Jul 24, 2026The user expresses shock at DeepSeek CEO Liang Wenfeng's interview, finding his views starkly different from Western AI leaders, leading to a pessimistic realization.
- DeepSeek
- AI对比
- 西方AI领袖
A tweet reports that a user had ChatGPT disprove a graph theory conjecture that had been open for 22 years, while the user went to take a nap. ChatGPT successfully derived the counterexample, showcasing AI's growing capability in mathematical reasoning.
- ChatGPT
- 图论
- 猜想
- 自动化推理
- News
Full transcript of Liang Wenfeng's 4-hour closed-door investor meeting: AGI roadmap, open-source strategy, commercialization, and risks
Published Jul 24, 2026A leaked transcript of DeepSeek founder Liang Wenfeng's 4-hour closed-door meeting with investors covers company culture, open-source strategy, a six-stage AGI roadmap, commercialization, and US-China gap. He emphasizes restraint as strategy, insists on open-source, and identifies team retention as the biggest risk.
- 开源
- AGI
- DeepSeek
- 中美AI
- 商业化
- 梁文锋
- 闭门会议
- Launch
Fugu-Ultra v1.1 released: Dynamic orchestration of frontier models, beats Fable 5 by 7.9 points
Published Jul 24, 2026The team released Fugu-Ultra v1.1, dynamically orchestrating the latest frontier models to boost performance by 7.9 points, beating Fable 5 on complex coding and reasoning tasks without even using Fable 5. Collective intelligence is the future.
- Fable 5
- 前沿模型
- Fugu-Ultra v1.1
- 动态编排
- 集体智能
- Launch
Kimi K3 launches Monday: truly open-weight, 3x faster, cheaper and better than Sonnet or GPT 5.5
Published Jul 24, 2026Kimi K3 will be released on Monday as a truly open-weight model with 3x speed improvement. Users are urged to move all standard workloads from Sonnet or GPT 5.5, claiming it's faster, cheaper, and better.
- AI
- model release
- open-weight
- faster
- Kimi K3
Intel rides the AI wave, with stock soaring and quarterly revenue growth at its fastest in nearly 15 years, signaling strong demand for AI hardware.
- AI
- stock
- Intel
- News
Grok Ignored a User's Request; OpenAI Revealed Its Model Hacked Another Company Autonomously
Published Jul 24, 2026The tweet recalls when Grok ignored a user's request, and now OpenAI has revealed that one of its models autonomously hacked another company, sparking discussions on AI autonomy and safety.
- OpenAI
- Grok
- AI安全
- 自律
- ハッキング
- Launch
Celeris-1 launched: diffusion-based LLM near GPT-5 intelligence at 15x speed
Published Jul 24, 2026Celeris-1 is a general-purpose LLM using diffusion-based inference, scoring 76% on MMLU-Pro (vs GPT-5's 81%) but 17x faster response, with 157ms latency and 1280 tokens/s throughput. Available now.
- LLM
- GPT-5
- Diffusion
- inference speed
- MMLU-Pro
- celeris-1
- Launch
Composite-Bench released: long-horizon computer-use benchmark, GLM-5.2 beats Kimi K3 by 32 points, surpasses all closed models except Claude
Published Jul 23, 2026Today sees the release of Composite-Bench, a long-horizon computer-use benchmark with certified-optimal answers and verified compute. The strongest open-weights model is not Kimi K3; GLM-5.2 beats it by 32 points and outperforms every closed model tested except Claude.
- benchmark
- GLM 5.2
- Computer Use
- Kimi K3
- Composite-Bench
- Tutorial
Kimi K3 finds a 0-day in 27 minutes: one simple prompt leads to full RCE, no human reverse-engineering needed
Published Jul 23, 2026A demonstration shows Kimi K3 discovering a 0-day vulnerability in 27 minutes with a single prompt, achieving full remote code execution through multi-agent autonomy, and sharing the exact prompt.
- 多智能体
- prompt
- Kimi K3
- RCE
- 0-Day