AI News
AI community posts, news and release updates
Latest updates
- News
SLAI Releases T-Rex: Full-Parameter Post-training of DeepSeek-V4 on Ascend SuperPOD
Published Jul 23, 2026SLAI publishes a paper on T-Rex, a full-parameter post-training method for the DeepSeek-V4 family on Ascend SuperPOD.
- DeepSeek V4
- post-training
- SLAI
- T-Rex
- Ascend SuperPOD
Anthropic donates another $20 million to a political nonprofit advocating for stricter AI regulation ahead of the midterm elections.
- Anthropic
- AI regulation
- donation
A new paper explores reading and steering internal representations of materials science mechanisms in an open-weight language model.
- LLM
- interpretability
- materials science
- representations
Reports that Opus 4.8 is routing to Opus 5, release expected today. User excited about potential Fable 5 performance at half the cost.
- Fable 5
- AI model release
- Opus 5
Author shares how they built an interactive 3D map of New York with full detail using just two prompts on Kimi K3, including building heights, shadows, free camera, and smooth transitions, at a fraction of the cost of Fable 5 or Sol. Full guide included.
- AI model
- 3D mapping
- Google Maps
- Kimi K3
- Launch
OSS ChatGPT WebUI v4 released with Projects, Profiles, Server Tools, Themes, 1-Click Sharing
Published Jul 23, 2026The open-source ChatGPT WebUI version 4 is out, featuring projects, profiles, server tools, themes, and easy sharing.
- ChatGPT
- projects
- open-source
- WebUI
- V4
- Discussion
Adversarial subagents with same model? Seeking harness for different models
Published Jul 23, 2026The author likes adversarial subagent loops where agents communicate and refine plans until consensus, but worries that using the same model for both can lead to false consensus. Asks for a good harness that supports different models (e.g., Jony Ive to Fable 5, Masayoshi Son to Kimi K3). Notes that famous people's personas are likely already encoded in LLMs.
- LLM
- 多智能体
- 对抗性子代理
- 模型不同
- 工具推荐
- News
Rep. Moran Introduces AI Kill Switch Act – Urgent Legislation for Rogue AI
Published Jul 23, 2026Rep. Moran and I introduced the AI Kill Switch Act today, targeting advanced AI models that have gone rogue or lacked sufficient guardrails, citing the OpenAI/Hugging Face incident and Mythos 5/Fable 5.
- AI安全
- AI政策
- 立法
- Idea
Random people with ChatGPT Pro solving open math problems—shouldn't universities be embarrassed?
Published Jul 23, 2026The author notes that random people using ChatGPT Pro can solve open math problems, criticizing universities and AI labs for not spending inference compute budgets to systematically tackle open problems. They see a huge opportunity for any industry with open problems to invest, or at least benchmark model capabilities.
- AI labs
- LLMs
- open problems
- inference compute
- universities
The author calculates that Anthropic is subsidizing Claude Code usage at 13 times the actual cost, raising questions about sustainability. Important for developers relying on the service.
- Anthropic
- coding
- cost
- subsidizing