Right but to detect close-enough spellings and word orders, using a curated index or catalogue of accepted examples, is one thing.

To train layers of algorithms in layers of machines on massive datasets to come up with close enoughs would be that but many times over the costs.

You would be a moron to use llms for spellchecking.

To clarify to you, not all programs are equal. Its not all different methods to do the same thing at the same cost.