According to a recent article from The Information, OpenAI has been demonstrating to some of its clients a new multimodal AI model that can recognize items and converse with you. The outlet claims to have seen it from anonymous sources and speculates that this may be a preview of what the corporation will unveil on Monday.
According to reports, the new model can comprehend images and sounds faster and more accurately than its separate transcription and text-to-speech models. The Information claims that the model “theoretically” can assist children with math or translate real-world signs, and that it would be able to assist customer service representatives in “better understanding the intonation of callers’ voices or whether they’re being sarcastic.”
The outlet’s sources suppose the mannequin can outdo GPT-4 Turbo at “answering some forms of questions,” but is easy at possibility of confidently getting issues inferior.
It’s likely OpenAI is moreover readying a brand modern built-in ChatGPT ability to to find phone calls, per Developer Ananay Arora, who posted the above screenshot of name-linked code. Arora moreover seen proof that OpenAI had provisioned servers supposed for proper-time audio and video verbal change.
None of this might be GPT-5, if it’s being unveiled subsequent week. CEO Sam Altman has explicitly denied that its upcoming announcement has one thing to make with the mannequin that’s presupposed to be “materially better” than GPT-4. The Data writes GPT-5 will most definitely be publicly launched by the stop of the year.
Altman moreover stated the company isn’t announcing a brand modern AI-powered search engine. But if what The Data experiences is what’s printed, it’d easy rob some wind out of Google’s I/O developer conference sails. Google has been sorting out the exercise of AI to to find phone calls. And definitely one of its rumored projects is a multimodal Google Assistant replacement called “Pixie” that might be aware at objects thru a tool’s digicam and make issues like give directions to places to purchase them or offer instructions on how to make exercise of them.
Whatever OpenAI plans to unveil, it plans to make so through livestream on its build on Monday at 10AM PT / 1PM ET.
