Practical use cases of LLMs in development projects

One of our recent episodes of The Puzzle Podcast with Bruno Pešec strayed off the beaten path somewhat, by exploring various use cases that I have found matching and respecting the current capabilities of LLMs such as ChatGPT (or self-hosted ones, such as WizardLM, LLaMa 2, Mistral-7b and Mixtral-8x7b) in product development projects.

Here is a non-exhaustive list:

• “Junior copywriter” parsing data and converting it into specific templates; e.g. technical specification language into features and benefits, rough ideas into value-proposition “adlibs”, AIDA-style mini-pitches.

• “Kickstart helper” generating something that might make sense as a starting point for your own copywriting or research.

• “Procrastination deleter” removing the psychological hurdle of not even knowing where to get started on a new topic. Just ask the LLM something like this: “I’m trying to wrap my head around topic X. Please provide an overview of the topic, and which of its elements would serve as a good starting point”.

• “Logic analyzer” (not the electronics one) detecting flaws in logic, fallacies, discontinuities in an argumentation, etc.

• “Example generator” for command-line software tools: for example, all the Puzzle Podcast videos are generated by using ffmpeg with command-line arguments that I defined after iterating with ChatGPT.

• “Condescension avoider” for getting started on a topic without having to deal with snooty self-appointed gatekeepers on online forums, where asking basic-yet-important questions might result in derision or elitism (and wasted time), instead of helpful pointers.

• “Conceptual rubberducking” for exploring a fuzzy topic from multiple angles in a longer discussion; such as determining factors that impact the segmentation of a market in terms of each segment’s JTBD complexity. In fact, this proved useful during my research on rental real-estate property management during the early stages of Breek.gr .

• “Framework generator / junior consultant” for generating a list of factors/parameters for a framework, and then using this framework to generate example combinations situations matching the framework’s parameters. After all, from my interview experience with McKinsey back in 2013, the understanding of creativity for those companies boils down to generation and enumeration of options based on an existing framework.

All this assumes that you never take what an LLM gives you at face value as a source of knowledge, but as an impulse for further research and discussion with those who actually know their stuff (see also “On using LLMs like ChatGPT to get things done” ).

Knowledge work will never be the same, but hey–knowledge work is never the same, ever. LLMs are merely yet another tool. We’ll see how far this tool, since something else will certainly succeed it.

“The tool itself is not going to steal your job or make you useless. Your lack of wanting to learn more, the resistance to learn more things through exploration is probably what’s going to make anyone less useful going forward–and LLMs are a good antidote to that. If you’re naturally curious, [learning to use LLMs] is amazing.”