nineteenth September 2022
Humanoid robotic mixed with GPT-3
Ameca, a extremely practical android, has now been upgraded to incorporate GPT-3, one of many largest neural networks and language prediction fashions.
Again in December 2021, UK-based Engineered Arts revealed what it described as “probably the most superior android ever constructed” – a machine with strikingly lifelike motions and facial expressions. Since then, the corporate has been working to improve Ameca (as she is named) with speech and different capabilities.
Within the video demonstration beneath, automated voice recognition has been mixed with GPT-3, a big neural community and language prediction mannequin that makes use of 175 billion parameters. This enables Ameca to recognise what individuals are saying and reply to questions. Earlier than talking, her output is fed to a web based text-to-speech service, which generates the voice and visemes for lip sync timing.
“Nothing on this video is pre-scripted – the mannequin is given a fundamental immediate describing Ameca, giving the robotic an outline of self – it is pure AI,” in response to Engineered Arts. “The pauses are the time lag for processing the speech enter, producing the reply and processing the textual content again into speech.”
