OpenAI's new GPT-5.4 clobbers humans on pro-level work in tests - by 83%

GPT-5.4 is also more reliable, producing 18% fewer errors and 33% fewer false claims than GPT-5.2, according to OpenAI.

Microsoft CEO says 'I look at all agents as users', and that means AI ones too, as each agentic bot 'will have its own identity'

The background to all this is Microsoft's claim that "Windows is evolving into an agentic OS". The implication is that there will be elements of the OS that give Copilot "agents" autonomy to perform ...