LLMs are a probability based calculation, so it will always skim to some degree, and always guess to some degree, and often pick the best choice available to it even though it might not be the best.

For folks who this seems elusive for, it's worth learning how the internals actually work, helps a great deal in how to structure things in general, and then over time as the parent comment said, specifically for individual cases.