The need
Many students learn AI from media coverage and intuition, not from the documentation, so they treat it as magic. They retry the same prompt until the result gets close, and they use one assistant for every task. These habits form during their studies and follow them into their first jobs, where errors have real costs.
What I owned
The proposal itself: the concept, the messages, the objectives and the three cases. This session has not been delivered yet.
The solution
Two core messages: do not use AI blindly, and do not use the same assistant for every task. Each case follows the same arc, where an ordinary user gets frustrated and an engineer reads the documentation, understands the system and gets the result.
Key modules
Case 1: diffusion models
Knowing that these models were trained on detailed captions tells you how to prompt them. Understanding how a tool works tells you how to use it.
Case 2: a Lebanese law assistant
A general assistant may cite articles you cannot check, while a dedicated tool connected through MCP returns the source text. Learn what a tool can connect to before trusting it on a high-stakes question.
Case 3: datasheet to NotebookLM
As a long chat grows, earlier content can drop out of context. Recognising that limit means moving to a tool that retrieves from the file for every question and shows citations.