When you’re outsource your thinking to generative AI, you diminish your own learning and the formation of a mental model. They usually give you the deltas to a system, those aren’t what matter most for model formation. The reason why the deltas take that specific form is.
Context is king
Both for you and the LLM. The AI has the context window, but you do not.
This is very clear when you work on complex systems like a codebase. You might read all the changes that are proposed, shape the deliverable and work with the LLM to get your desired output. But since this administrative friction and not constructive friction you’re missing something that the LLM isn’t: context.
Every consideration, option and the edges of the system are considered by the AI, that you as the human in the loop will not see or won’t even come into contact with, when you miss this context your mental model will have gaps. This is not new, as the same thing happens when you work without an LLM and don’t come into contact with the edges of a system, but since you’re in the driver seat, you will have more context than when you’re not.
Eggplant in the fridge
When you’re only aware that you put an eggplant in the fridge (a code change to the system). You know that there is an eggplant in the fridge, but you’re not sure why.
Is this eggplant there because you’re restocking, or are you working up to a recipe? Perhaps you’re replacing a mouldy vegetable or it wasn’t supposed to be there anyway.
You don’t even know what’s in the kitchen.
You’re missing the context that the LLM took.