GPT-4 provided related capabilities, giving customers a number of methods to work together with OpenAI’s AI choices. But it surely siloed them in separate fashions, resulting in longer response occasions and presumably increased computing prices. GPT-4o has now merged these capabilities right into a single mannequin, which Murati referred to as an “omnimodel.” Meaning quicker responses and smoother transitions between duties, she stated.
The outcome, the corporate’s demonstration suggests, is a conversational assistant a lot within the vein of Siri or Alexa however able to fielding rather more complicated prompts.
“We’re taking a look at the way forward for interplay between ourselves and the machines,” Murati stated of the demo. “We predict that GPT-4o is admittedly shifting that paradigm into the way forward for collaboration, the place this interplay turns into rather more pure.”
Barret Zoph and Mark Chen, each researchers at OpenAI, walked by way of a lot of functions for the brand new mannequin. Most spectacular was its facility with dwell dialog. You could possibly interrupt the mannequin throughout its responses, and it will cease, pay attention, and modify course.
OpenAI confirmed off the power to alter the mannequin’s tone, too. Chen requested the mannequin to learn a bedtime story “about robots and love,” shortly leaping in to demand a extra dramatic voice. The mannequin acquired progressively extra theatrical till Murati demanded that it pivot shortly to a convincing robotic voice (which it excelled at). Whereas there have been predictably some brief pauses through the dialog whereas the mannequin reasoned by way of what to say subsequent, it stood out as a remarkably naturally paced AI dialog.