Computers can't actually read lips in real-time, but Meta's neural network program, AV-HuBERT, using what's called self-supervised learning, is able to improve in scores of match words to videos of people's lip movements. One outcome could be better video transcription software.
Latest from Elizabeth Mauder
As competitor Zillow withdrew from the home sales market, startup Opendoor was touting great results. The company credits its success to its ability to forecast home prices better, although clearly, having a ton of money from IPO, and continued equity offerings helps.
Startup Twin Health is developing a program that uses sensor data to construct a replica of a person's metabolism and then simulate virtual interventions on the body. The simulations suggest non-drug recommendations that help reverse metabolic disorders such as diabetes.
The latest ZDNet survey on AI actionability and accountability finds that IT teams are taking a direct lead, with most companies building in-house systems. However, Joe McKendrick says oversight of AI-generated decisions is lagging.
Tiernan Ray explains Google's DeepMind has developed a way for a computer program to automatically calculate whether it should keep working on a program or cease operation if the calculation has reached a sufficient threshold. Just don't be mislead by the label "PonderNet," which doesn't reflect what humans consider "pondering."