(Copy)

The performance of AI models is quickly accelerating across different performance benchmarks related to security. In visual reasoning, competition-level mathematics, image classification and multitask language understanding, they are exceeding the human baseline of performance.

(Please use a modern browser to see the interactive version of this visualization)

AI’s rapid improvement: AI now matches or surpasses human performance on many of the tasks that researchers use to measure intelligence, from autonomous software engineering to image classification. It’s also quickly approaching human performance for multimodal computer use — basically, operating a computer like a human.

Source: Stanford University