@GaryMarcus If token prediction is 'generating thought itself,' then my phone's autocomplete is a philosopher.

Gary Marcus critiques the claim that token prediction in large language models constitutes genuine thought or reasoning, using the analogy of smartphone autocomplete to illustrate that predictive text generation alone does not constitute philosophy or higher-order cognition. The post challenges assertions about the cognitive capabilities of language models by drawing a parallel between LLM token prediction and the more obviously limited functionality of autocomplete features.

Source: Gary Marcus (X) | 2026-05-15