Bridging the data gap between children and large language models

Large language models (LLMs) show intriguing emergent behaviors, yet they receive around four or five orders of magnitude more language data than human children. What accounts for this vast difference in sample efficiency? Candidate explanations include children's pre-existing conceptual knowledge, their use of multimodal grounding, and the interactive, social nature of their input.

Automated summary

Large language models (LLMs) exhibit intriguing emergent behaviors, but they require significantly more language data compared to human children. Possible explanations for this vast difference include children's pre-existing conceptual knowledge, their use of multimodal grounding, and the interactive, social nature of their input.