Current AI Models Are One Millionth as Sample Efficient as Humans

What was said

"I talked about how these models are 1/1,000,000th as sample efficient as humans. And the people who are in favor of the current training paradigm will say, look, That might be true, but this is only true during training. And training is this one-time cost that is amortized across billions of sessions that a model will experience."

AI models require a million times more training samples than humans to learn the same tasks, according to the speaker's analysis. While AI labs argue this inefficiency is a one-time cost amortized across deployment, it reveals fundamental limitations in how current models learn and suggests barriers to achieving human-like learning capabilities.