Keep three layers separate: AI is a broad field; a large language model is a type of model that processes and generates language; a chat interface is one way a product lets you use a model. A feature in one chat product is not automatically a feature of every model.
Ask what a task needs
List the input, desired output, and a way to check it. For a note summary, the input is your notes, the output is a short summary, and the check is whether each statement matches the original. This is more practical than asking whether a model is simply smart.
Context is not unlimited memory
Instructions and material in a conversation affect the response. Do not assume every earlier detail will remain available. Restate essential background and constraints when starting a new task.
Fluency is not correctness
Language models can produce plausible but inaccurate answers. Treat an answer as a candidate result and verify factual claims against their sources. Confidence in the wording is not evidence.
Try it
Explain the difference between a model, a chat product, and a task in your own words. Then define the input, output, and acceptance criteria for a familiar task. You are done when the task is clear without naming a tool.