Beff Jezos Contrasts Mode-Covering LLMs With Mode-Seeking Brains

Beff Jezos posted on X that AI models are mode-covering while real brains are mode-seeking. The message states that LLMs are trained to minimize cross entropy. It adds that brains actually minimize variational free energy. Jezos wrote that stochastic-thermodynamic algorithms and computation provide the path to a post-slop world. The post includes a photo attachment and comes from the account of the Extropic AI founder and e/acc proponent.