LLMs and the Brain: Autoregressive Mechanisms of Human Cognition
Prof. Elan Barenholtz delivered a keynote on how key computational features of transformers — embeddings, attention weighting, and autoregressive updating — align with known neurophysiological signatures including temporal decay, graded residual activation, and recurrent cortical loops. The lecture demonstrated how transformer-style architectures may offer a unified computational perspective on human thought and behavior.