Even Babies Are Still Way Better at Learning Than AI Models

Despite massive advances in AI, large language models are far less efficient at learning language than human babies. Toddlers can form coherent sentences after absorbing just 10 to 30 million words, while training GPT-2 on the same amount produces nonsense. Experts say AI still requires enormous datasets to reach basic proficiency, making this inefficiency a major challenge for companies pursuing human-level AI.