So OpenAI launched Operator (as did Anthropic with “Computer Use” and Google with “Project Mariner” some time before): What does it mean?

It is a large-vision-language-model-backed AI assistant that can to some degree use your computer via the normal graphical user interface to solve certain tasks.

Such personal AI assistants that really deserve this name (by being able to help with most things that can be done with a computer) will be the thing in the next 2-3 years. Me and others have been pointing this out in talks since 2023, so this is expected.

What we currently see are the first prototypes of such products, the early days so to says: Not yet very capable, in many aspects very limited, but this will improve. The underlying tech ist still fundamentally the same we know from the current chat-based assistants (large language and vision-language models with agentic workflows for “chains of thought”), hence they inherit the same fundamental limitations (hallucinations, energy consumption, no actionable common sense). So: No AGI or anything like this. But it will get very useful.

Link to original article