OpenAI is preparing to launch an AI agent that can do your work for
you
OpenAI is nearing the launch of "Operator", their first autonomous AI agent.
OpenAI rumor season is continuing strong.
The report this time: OpenAI is preparing to launch an AI agent codenamed “Operator” in January.
The agent would be able to use your computer on your behalf. So, theoretically, it could write code, book restaurant reservations, buy plane tickets, and send emails.
Although the report got a lot of buzz, it didn’t come completely out of the blue. During OpenAI’s Reddit AMA, CEO Sam Altman called agents “the next giant breakthrough.”
It’s hard to say he’s wrong. By OpenAI’s own internal metrics, agents are the next level in the staircase to artificial general intelligence (AGI).
So, if “Operator” is truly coming, it’d be a huge burst of confidence for the AI researchers currently stressing about the “scaling law” breaking down. In other words, AI isn’t improving as quickly with more data as it used to.
Agents, at least temporarily, would remove that concern. And, in the most optimistic scenario, be AI’s next major breakthrough.
Some part of me hopes AI retains its status as a companion to humans, not become super-human. I'd love for it to do some mundane work for me, as it does already, but I would personally prefer it continue to be limited to that role. Still interesting to see the future unfold and perhaps, if super-intelligence comes to fruition, it may not be as frightening as I imagine.
First Anthropic's recently released Claude with Computer Use, then Open AI. It's interesting to witness multiple companies pursuing AGI by advancing in this direction. Perhaps this sort of RLHF through temporal human interaction data is what the AI requires to break through its current plateau.