@orgo

Computer-use agents are still in the early stages. However, all major foundation models - such as GPT, Claude, and Gemini - can now look at a screen and operate a computer like a human. Models are improving rapidly: the best models recently scored 63% on the OS World benchmark, up from 20% at the start of this year. For reference, the average human scores around 70%. At this rate, AI will surpass average human computer proficiency by the end of next year.

1 reply

0 recast

0 reaction