New EgoBabyVLM Benchmark Tests Whether AI Can Learn Like an Infant From Head-Camera Video

Researchers from Meta FAIR, Stanford University, the University of Tokyo, and École Normale Supérieure have introduced a new benchmark called the EgoBabyVLM Challenge, which measures how well vision-language models can learn from the noisy, real-world video that infants actually see—and finds that today's leading systems still fall short. The effort reflects a growing conviction among researchers that the architecture of a baby's brain may hold clues for building far more efficient artificial intelligence.**