LLM failure modes map surprisingly well onto ADHD cognitive science. Six parallels from independent research.
I have ADHD and I've been pair programming with LLMs for a while now. At some point I realized the way they fail felt weirdly familiar. Confidently making stuff up, losing context mid conversation, brilliant lateral connections then botching basic sequential logic. That's just... my Tuesday. So I went into the cognitive science literature. Found six parallels backed by independent research groups who weren't even looking at this connection. Associative processing. In ADHD the Default Mode Network bleeds into task-positive networks (Castellanos et al., JAMA Psychiatry). Transformer attention computes weighted associations across all tokens with no strong relevance gate. Both are association machines with high creative connectivity and random irrelevant intrusions. Confabulation. Adults ...
