Why kids outperform LLMs using 100,000x less data
mit researchers identified a 'data efficiency gap' where children outlearn ai with significantly less data, suggesting that reverse-engineering human cognition could yield more efficient models.
mit technology review reported on the 'data efficiency gap,' noting that llms require vast datasets while children achieve fluency with minimal exposure. researchers hope that understanding this human learning mechanism will lead to ai architectures that require less training data. this line of inquiry seeks to solve fundamental challenges in ai efficiency and answer enduring questions about cognitive science.