Comment by Hedepig

The static nature of LLMs is a compelling argument against the reasoning ability.

But, it can learn, albeit in a limited way, using the context. Though to my knowledge that doesn't scale well.