BREAKING
EgoBabyVLM Tests AI vs. Infants
Built by Meta FAIR and Partners
0
hrs
BabyView recordings
0
+
DevBench trials
How the Benchmark Works
1
Egocentric video
↓
2
Object recognition
↓
3
Grammar & semantics
↓
4
Score grounding
Scaling Data vs. Learning Like a Baby
Big Models
Today
●
Vast data and energy
●
Weak physical grounding
Infant Learning
Goal
●
Limited first-person input
●
Embodied common sense
A Yardstick, Not a Solution
AI NEWS BLITZ
A new benchmark asks whether AI can learn from the world like a baby does.