Current LLMs - Width is high, but depth lacks any understanding, logic and intuition

A user tasked an LLM, Astra, with implementing details for a private Python pipeline, involving 300 points, each taking 1-2 hours. While Astra provided feedback on each implemented point and mentioned numerous tests, the user concluded that current LLMs, despite their breadth, lack true understanding, logic, and intuition. This observation led to a sense of relief that human intelligence remains valuable, with the only perceived threat being a "paper clip scenario."

This user's direct experience with Astra, involving 300 detailed tasks, offers a specific, large-scale example of current LLM limitations, unlike general discussions of AI capabilities.