In 2026, researchers from Stanford, MIT and several universities submitted the most advanced models to videos filmed by young children. The machines had to understand the world from this raw flow, but they failed heavily. The test of the AI against the baby shows that the toddlers learn even faster, with far less examples. A thousand hours of video seen by babies The test, called EgoBabyVLM, comes from a simple idea. Cameras fixed on the heads o…
This story is only covered by news sources that have yet to be evaluated by the independent media monitoring agencies we use to assess the quality and reliability of news outlets on our platform. Learn more here.
In 2026, researchers from Stanford, MIT and several universities submitted the most advanced models to videos filmed by young children. The machines had to understand the world from this raw flow, but they failed heavily. The test of the AI against the baby shows that the toddlers learn even faster, with far less examples. A thousand hours of video seen by babies The test, called EgoBabyVLM, comes from a simple idea. Cameras fixed on the heads o…