Comment by kelseyfrog

AI assisted boardgame design process.

I enjoy exploring AI-assisted process development. Things I've learned are:

* Building nôn pipelines for generating placeholder art. GPT-5 excels at writing code nodes.

* LLM for LaTeX generation (note: GPT-5 generally fails at even moderately complex LaTeX and often adds complexity while endlessly orbiting the problem without solving it. On of the other hand, it's failed solutions eventually give me enough datapoints to develop the correct solution on my own. Probably faster than learning enough LaTeX to develop them on my own. Sadly I can fix broken LaTeX much easier than writing correct LaTeX.

* Using LaTeX to read card data out of CSVs is difficult to get right (esp with embedded LaTeX) but honestly pretty good for rapid prototyping.

* Searching games for mechanics to solve specific interaction problems is an ok-ish problem for GPT-5. I wish I had a vector DB of indexed PDF manuals for the top 5k bgg games, but that's too big of a problem to solve.

It's incredibly cheap and easy to generate reasonable approximations of thematic text sufficient for a prototype.

Overall, I went from idea to first round of playtesting in about 2-3 weeks. Previous similar attempts had been 2-3 months. I feel quite a bit less attached to darlings, but the downside is a bit of shame attached to having playtesters work around errors originating from AI usage rather than errors generated by my own oversights