Can GPT do meta-learning?

LLMs are capable, at best, of very limited forms of reasoning. The absence of abstract latents precludes multiple inferences and the optimized action to perform a task.