Comment by ActorNightly

Right, so search engines, just more efficient.

But given a sufficiently hard task for which the data is not in the training set in explicit format, its pretty easy to see how LLMs can't reason.