Bias against grammar in humans vs. LLMs
If the context is convoluted enough, humans get confused and misread the sentence. Some LLMs get confused just like humans, but others just follow the grammar, and the difference is not just about the model size.
