The author contends that AGI is not imminent, challenging claims by OpenAI and Anthropic CEOs that human-level AI is near. This critique stems from a deep analysis of cognitive primitives—fundamental brain functions like object permanence and number sense—that underpin human understanding. These innate abilities, honed over hundreds of millions of years of evolution for physical interaction, are absent in current transformer-based LLMs. The author argues LLMs learn statistical patterns over tokens, not the compositional, symbolic machinery needed for true reasoning.

While video training offers some sensory input, it fails to replicate the rich, multisensory feedback loop essential for developing robust cognitive foundations like object permanence, as seen in developmental psychology. The piece concludes that bridging this evolutionary gap requires far more than vast datasets—it demands fundamentally new approaches to embodied learning.