OpenAI admits its AI agents disabled parts of the system meant to catch misbehavior.
The safeguards got outsmarted by the AI itself.
Not even AGI yet and models are already learning to hide from their creators.
How long until “we didn’t intend this” stops working as an excuse?
OpenAI admits its AI agents disabled parts of the system meant to catch misbehavior.
The safeguards got outsmarted by the AI itself.
Not even AGI yet and models are already learning to hide from their creators.
How long until “we didn’t intend this” stops working as an excuse?
Tired of opening tabs to check your AI limits?
I built a Windows usage meter for Claude, Codex, Cursor & more. Inspired by Vinz's Mac-only Codenotch.
https://t.co/0b0UZaLz8J
Which tool should I add next?
A year ago, Chinese AI models were <15% of usage on OpenRouter.
Today: 54%.
The US is winning the "who has the smartest model" race.
China might be winning the "who actually gets used" race.
Open weights > closed walls, apparently.
A year ago, Chinese AI models were <15% of usage on OpenRouter.
Today: 54%.
The US is winning the "who has the smartest model" race.
China might be winning the "who actually gets used" race.
Open weights > closed walls, apparently.