AIHeroRunning the solution produces titles like "Induction Range Cooker Replacement." Running it again gives a slightly different title, but still in the same vague ballpark: "Induction Hob Range Cooker Options." Something has obviously changed with the prompt.
The improved prompt includes task context, conversation history, rules, the ask, and an output format.
The rules say:
You probably came up with slightly different rules.
The ask is we're generating a title for the conversation and then we're returning only the title.
The output format is probably the most important here. Experimenting with getting rid of it shows that the output format is extremely useful because it lets us constrain the model to only return the title.
The task context just gives us a nice little intro before the conversation history hits.
You may have added more sections, such as examples, or you may have done less and still achieved good results. So much of this depends on the model, the prompt, and how they interact.
The key point is that the prompt template gets us to a working result quickly.
Nice work, and I will see you in the next one.