- OpenAI CEO Sam Altman described the ultimate AI app to the MIT Technology Review.
- He envisions a 'super-competent colleague,' a major upgrade from ChatGPT, which he called "dumb."
- Altman's vision is that AI will take on real-world tasks — not just function as a chatbot.
In Sam Altman's vision of the future, AI is a little intimidating.
"What you really want," the OpenAI CEO told the MIT Technology Review, is a "super-competent colleague that knows absolutely everything about my whole life, every email, every conversation I've ever had, but doesn't feel like an extension."
And they're self-starters that don't need constant direction. They'll tackle some tasks, presumably simpler ones, instantly, Altman said. They'll make a first pass at more complex tasks, and come back to the user if they have questions.
The bottom line is that Altman wants AI to function as more than just a chatbot. It should help people accomplish things in the real world, he said.
That would be a massive step up from what OpenAI offers right now.
Altman reportedly referred to ChatGPT as "incredibly dumb" even though workers are already using it to accelerate their workflows, develop code, write emails, and more. So, there's no telling how much more productive we'll get once Altman's magical model colleague hits the market.
Altman didn't specify when this tool will be available and how advanced AI must be to support it. The company's other offerings, like the video generator, Sora, and image generator, DALL-E, still require considerable guidance to complete tasks. They also aren't designed to perceive information from the environment and use it to achieve specific goals.
But OpenAI's forthcoming language model, GPT-5, might be a step in that direction.
A source who's seen it previously told BI it was "materially better" than existing models. The source also said that OpenAI is developing a service where users could call an AI agent to perform tasks autonomously.
Sources have said GPT-5 might be out mid-year. Altman, however, isn't saying much.
"Yes," he simply told reporters this week at an event in Cambridge, Massachusetts, where he was asked when OpenAI would release GPT-5.