What's different about Riddler compared to _most_ casual word games is that it's designed to generate its own puzzles and content autonomously with LLMs.

I say _most_ casual word games because there is a brilliant exception – Wordle. [https://www.nytimes.com/games/wordle/index.html]

The genius of Wordle lies in its simplicity. Players have six attempts to guess a five-letter word, receiving feedback for each guess in the form of colored tiles that indicate when letters match or occupy the correct position. It can be considered autonomous in the sense that (in theory), each daily puzzle can be created easily by sampling a five-letter word from a set dictionary.

But is Wordle truly autonomous? Almost.

The NYTimes caught some flack in 2022 when a daily Wordle answer – FETUS – was sampled at an inopportune time. The Roe v Wade supreme court draft ruling had just been leaked, and the 5-letter word was poorly received by enough NYT Games players for the Times to make a statement: "Today, some users may see an outdated answer for Wordle that seems closely connected to a major recent news event. This is entirely unintentional and a coincidence – today's original answer was loaded into Wordle last year." [https://www.theguardian.com/media/2022/may/09/new-york-times...]

NYT Games scrambled to refresh the word. If a single word – innocuous in a vacuum – can upset users in certain unforeseeable contexts, how does one ensure that longer form generative content is curated in a way that makes it both safe AND fun? This is a question I'm still trying to answer.

Riddler is clearly inspired by Wordle's simple design – I'm not a designer, so I wanted to keep this somewhat constant for the time being. However, Riddler is more similar in gameplay to word ladders, invented in the 1800s and which happen to have a similar layout to Wordle [https://en.wikipedia.org/wiki/Word_ladder]. So some aspects of the Wordle layout work.

Currently, every Riddler player gets the same daily puzzle, which makes sense for competitive and sharing reasons. I'm currently reviewing the puzzles myself, using scripts to regenerate certain riddles and images if necessary. I think this too can be automated through meta-prompting and LLM self-review of output. So is Riddler autonomous? Almost.

Riddler could allow users to craft their own puzzles on demand, which I think is interesting. Riddler's backend uses path finding algorithms to find paths between two words, but I've found that the most efficient path between two words is not ideal for gameplay because this leads to repetitive puzzles given a static dictionary. Dictionary curation is quite difficult, by the way, and I plan to use LLMs for helping to improve on this front as well.

I'm a server engineer, not a game designer, so I'd really appreciate any feedback on the gameplay since it's not quite there yet. I want to continue creating games like this – and have a dream to move to other types of curated, generative content like interactive articles, podcasts, and videos.

Interested in collaborating or discussing opportunities? I'm looking for something new! Reach out at po (at) po [dot] studio.

Thank you!