It's a pretty cool proof of concept. I did run into some weirdness with it right away when trying the Harry Potter book. The app kept switching between 2nd and 3rd person for Harry.
The main thing for me, though, was the feeling of emptiness I got while playing.
I love text adventures, having grown up with Infocom games. The thing about them is that you can feel the choices made by the writer / programmer, just like you can feel the human author behind a book.
I'm sure part of what feels empty to me is because I know it was autogenerated. But I feel that even if I was shown this without knowing an LLM was behind it, the gameplay wouldn't be as rewarding as something written by a human using Inform.
This is true for me as well. Once you realise that with the right prompts you can get the LLM to go off on any tangent you want, it starts to ring really hollow. Add to that the inability of LLMs to maintain world simulations of any fidelity at all, and it becomes difficult to motivate spending time with them in that way.
Freedom is valuable in games, especially text adventure games. And an LLM is a powerful way to give the player freedom, since it can respond to almost anything.
But for games to shine they also need carefully considered constraints - when the player bounces off constraints they start to understand how the game's world and systems function, which lets them build a mental model and be able to start thinking the same way the designer(s) thought and come up with solutions for puzzles or decide how to react to a given situation.
What makes a maze engaging as a challenge is that your path is closed off in some places and not in others. Ideally any maze also has one or more concrete solutions, so the player is rewarded for mastering the maze by finding an exit, or maybe finding objectives or creatures hidden at key locations in the maze.
You can probably use modern LLMs to construct this sort of world and set those constraints, but I wonder how far we are from being able to also have the LLM maintain that world state and enforce the constraints?
That's a very insightful take. I think an LLM would have some role in a good text adventure game, in terms of being able to understand any natural language input and respond intelligently. But as you point out, it would need to do so within strict constraints set by the game designer.
I haven't played with the most state of the art parsers that are available right now, so I wonder how large the gap is between a great parser and using an LLM to process user input.
The main thing for me, though, was the feeling of emptiness I got while playing. I love text adventures, having grown up with Infocom games. The thing about them is that you can feel the choices made by the writer / programmer, just like you can feel the human author behind a book.
I'm sure part of what feels empty to me is because I know it was autogenerated. But I feel that even if I was shown this without knowing an LLM was behind it, the gameplay wouldn't be as rewarding as something written by a human using Inform.