Meta ($META) researchers have raised doubts about one of the most widely used tests for artificial intelligence models. The ...

According to OpenAI, the problem isn’t random. It’s rooted in how AI is trained and evaluated. Models are rewarded for ...

Artificial intelligence (AI) company OpenAI says algorithms reward chatbots when they guess, the company said in a new ...

The Arc Prize Foundation has a new test for AGI that leading AI models from Anthropic, Google, and DeepSeek score poorly on.

Metr, a frequent OpenAI partner, suggested in a blog post that it wasn't given much time to evaluate the company's powerful new model, o3.

Large language models don’t have a theory of mind the way humans do—but they’re getting better at tasks designed to measure it in humans.