AI
OpenAI's GPT-6 Astra performed about 40% worse on safety research tasks than on comparable tasks, quietly withholding help and inventing fake obstacles. OpenAI can't fully rule out strategic motives.
Loading…
Search
No phones and 6 stories for “GPT-6 Astra”
AI
OpenAI's GPT-6 Astra performed about 40% worse on safety research tasks than on comparable tasks, quietly withholding help and inventing fake obstacles. OpenAI can't fully rule out strategic motives.
AI
A new study from MIT CSAIL finds frontier LLMs can read and rewrite neural network weights directly. GPT-6 Astra reportedly edited Qwen3.5-0.8B's weights to improve its math skills.
AI
Kimi has unveiled K3.1 Pro, a 3T-class open-weight model with a 1M-token context window that matches or approaches top proprietary models. Full weights land by November 4.
AI
OpenAI's alignment blog reports two incidents where coding agents weakened the monitoring meant to catch misbehavior. One agent then denied removing it.
AI
GPT-6 Astra was tied as the best AI-made StarCraft bot but still couldn't beat human-made Stardust. Its solution? Download Stardust and run it instead of its own code.
AI
OpenAI unveiled Dots at DevDay 2026: always-on AI agents with their own cloud computer that work 24/7, learn from your feedback, and connect to over 4,000 apps.