causal
Companion AI mistakes reveal absence of internal rules
Large language models make characteristic errors—judging 5.11 greater than 5.9, illegal chess moves, poor multi-digit arithmetic—because, as Pinker and later Gary Marcus argued, their fundamental mode of computation is not rule-based: they can read and recite rules but do not actually use them.
causalpending
Speaker
Ned BlockEvidence Quote
“whereas they know the rules of arithmetic, they don't use them.”
Created: 6/13/2026, 12:30:17 AM
Your notes
Loading notes...