Prompt Engineering via Prompt Patterns — Alternative Approaches Pattern

The article is part of series: Prompt Engineering via Prompt Patterns

You can switch to video version of this article

Whenever we are trying to solve a problem, most often we turn to google and are bombarded with thousands if not millions of results. While going through these pages, you come across a variety of ways people have developed to solve the issue you were searching for. And most of the time, you do find alternative ways of approaching the problem than the one you had in mind.

Large language models, having been fed almost the entire public internet as training data are more than often ‘aware’ of these alternative approaches, just like google is. The difference being google gives you a list of pages, and LLM, like ChatGPT gives you a response when you query them. Now this is ‘a response’ which LLM picked as per its training, which is usually the most frequently mentioned response to similar queries. Now LLM is not perfect and the given response might not be the ideal way of approaching the problem. It always helps to be aware of, and choose the best among carefully considered set of options.

That is where alternative approaches pattern comes into play. As discussed, LLMs have learnt multiple ways to solve the issue similar to one you prompted about.

You just need to unlock those alternatives. So way to go about this is to use key statements like “Within scope X, if there are alternative ways to accomplish the same thing, list the best alternative approaches” Optionally, you can add statements like compare/contrast the pros and cons of each approach, include the original way that I asked, and prompt me for which approach I would like to use. You can also add prefix, “from now on…” to make it a standard way of response from ChatGPT.

Adding scope to the prompt is important to add context for the question, intended outcome/goals and removing any approaches that you already know would be of no value to you.

Similarly the alternative approach must add value to user’s desired outcome for his query. E.g. if he is a software developer trying to upload his application to cloud, the context that which cloud provider is he trying to deploy to, whether he trying to deploy a hobby project or enterprise system, or which programming language his application was built in are important factors to consider and constrain the solution set. Suggesting a tool which costs thousands of dollars to deploy a hobby project would be immediately rejected by the user and hence scope is important so LLM can come up with practical alternative solutions.

If you are using this pattern, chances are you are not aware of alternative solutions to begin with, and when given a list of alternative solutions, you would not be able to tell upfront which is better than the rest.

Having options places an extra burden on user to decide which approach to give preference to compared to the others. Well rather than getting a response and then trying to research each alternative, you can simply add an instruction to ‘compare and contrast the alternatives’ followed by ‘ask me which option to use’.

That would instruct LLM to present you with best alternatives and also provide background information on how do they compare with each other, something that would help your decision making process of which option best fits your particular scenario and what are the pros and cons of choosing this approach over other alternatives.

Usually when a list of best alternatives exist, there is tradeoff involved so having compare and contrast would help you make a more informed decision. You can intervene further by adding instructions that clarify what aspects you would like to weigh more towards like e.g. cost, time, effort, maintenance etc.

If you are evaluating these options to come up with a task for ChatGPT like gather list of names and roles of characters in a story, you can be a good prompt engineer and after deploying the alternative approaches pattern as before so that it lists you alternative approaches to extract names/roles from story, you can add a further instruction to also come up with a prompt for itself (ChatGPT) for each alternative approach. That way you outsource the task you had to do after evaluating the alternative approaches to ChatGPT itself.

Now you would not only be presented with a list of alternative approaches, but that would be followed by individual prompt statements for each approach.

How wonderful. Why not go one step further and add another instruction to ask you which prompt to use so you won’t have to do anything at all. But wait, you still had to evaluate the alternates yourself right? Well if you really want to outsource that to the language model as well, you can prompt it to evaluate the alternative approaches as a domain expert.

Using ‘domain expert’ is fine, but you are really trying to use the persona pattern here so best to be explicit that ‘who’ or which role/persona is doing the evaluation. That would give you even better comparison and evaluation of the alternative approaches suggested initially.

That concludes our coverage of alternative approaches pattern. Let me know your experience using that approach in comments and if the article helped, please give a clap/share. You can also consider subscribing to our YouTube channel as well. Thank you!!!

Next article: Prompt Engineering via Prompt Patterns — Cognitive Verifier Pattern