Penny Nichols, Troubleshooter

Every once in a while the question of “what is IF?” comes up, and I have a couple of stock answers: one is that “IF” is a community-based discourse rather than a genre, and another is that “IF” is whatever we IF people are playing and talking about (these are equivalent formulations, just with more or less pretension according to taste). The other one I tend to trot out draws more from how academic disciplines are functionally defined, and holds that anything one can usefully analyze via the approaches IF critics have developed counts as IF.

These are broad definitions by intention, but by that last one, it’s very hard to consider that Penny Nichols, Troubleshooter is IF. See, this isn’t a complete game; rather, it’s a prompt you can paste into an LLM in order for it to create a game-like experience for you. As such, it’s pointless to talk about the stuff I usually do when writing a review: there’s no pre-baked prose whose quality I can assess, no ending I can weigh for thematic resonance, no puzzles that might be more or less fair. It’s all just down to whatever the plagiarism-bot feels like spitting up in the moment – so given that, as well as what I think is a well-founded reluctance to use an LLM unnecessarily, I didn’t actually bother to “play” this game.

There are other approaches to game criticism than the ones prevalent in our little community, though, and given the format Penny Nichols uses, I couldn’t help considering how it would look through the lenses tabletop RPG reviewers use when looking at scenarios. Those folks tend to look at questions like “how well are the scenario’s theme and flavor communicated to the GM so they can run it as intended?”, “are there raw materials here to allow the GM to construct a well-paced adventure”, “are the mechanics well thought-out?”, and “how railroaded is this adventure likely to be in practice?” And these are questions one can ask of the Penny Nichols prompt.

Unfortunately I don’t think it comes out very well on any of them. The prompt is quite short and devoid of any consistent vibe; there’s an underbaked science-fantasy theme that provides some proper nouns but no coherent guidance to a human intelligence as to how to play it. Like, here’s what we/the LLM are told of Penny:

The player character is Agent Penny Nichols, an Insurance Investigator from the Solar Insurance Company on Mercury.

Hue 150 (Divination & Illusion specialist).
Prefers indirect investigation, including cover identities.
Member of the Circle Trigonist faction.
Does that “hue” thing indicate Penny can do magic? What’s a “Trigonist”? Is locating an insurance company on a planet that’s consistently so hot it radiates mostly as a black body an indication that there’s some fraud going on, or are people just dumb? Your guess is as good as mine (and much better than ChatGPT’s); this is slightly better than “make up some bullshit,” but not by much.

As for the “plot” of the scenario itself, there’s more concrete reason to think that tabletop RPG design is the best way to think about this since it explicitly says the story should proceed according to the four-act kishokentetsu structure that was all the rage in RPG circles like five years ago. But the implementation of the structure is incredibly sketchy, not even running to 200 words: basically, there’s a space station studying an artifact, but the artifact has vanished, so you’re sent in to investigate. There’s meant to be a mid-story twist where you can find out that the artifact was a hoax by the lead investigator, because he wanted to get more funding; but then the final twist reveals that the artifact (or the station itself, the prompt isn’t clear) is actually a dragon’s hoard (or maybe the dragon itself?) that created the lead investigator as a psychic projection, in order to get the attention on which dragons (and hoards?) subsist. The resolution requires the player to “contain, banish, or escape before [the dragon] consumes more” (there’s no mention anywhere of the dragon having previously consumed anything).

Look, I’ve run a bunch of tabletop RPG adventures, and not to put too fine a point on it, but this one sucks. Hell, the notes I scribble to myself for scenarios I’ve come up with and already live in my brain contain way more detail about the psychology of the characters, how to construct challenges that are engaging to deal with, ideas about how to manage pacing, and particular bits of dialogue or turns of phrase to incorporate in my narration. Speaking as a reasonably experienced GM, I’d find this prompt worse than useless: it doesn’t give me any of the stuff I’d look to a scenario to provide, and in the time it’d take me to read, understand, and attempt to spackle over the holes of this prompt, I could come up with something far better using only my own creativity.

So that’s my assessment of what the author submitted to the Comp as a “game”, but I was morbidly curious about what could be included in the “walkthrough” file, since of course there’s nothing to walk through. Turns out it’s some commands that (might?) work to complete the scenario under Claude.ai, as well as a sample transcript of the author “playing” the game with ChatGPT. And oh lord, as bad a mood as reading the prompt put me in the transcript was worse.

For one thing, ChatGPT seems to insist on presenting everything as bullet-pointed lists of information and options, with embedded emojis, meaning reading it feels like being trapped in an Axios article (What they’re saying: “this is literally hell,” according to Mike Russo), and also makes me wonder how the author reconciled the “you can type anything and the game will understand it!” promise of LLMs with the reality that it was providing an interface indistinguishable from that of an especially low-effort choice-based game. For another, while the blurb promises that Penny Nichols is a “Star-trek style away mission”, ChatGPT sure seemed to think it’s a high fantasy setting where all your actions involve casting magic spells. And actually the prompt in the transcript isn’t the same as the prompt in the Comp submission!

The transcript at least explains the last of these discrepancies; halfway down, the author realizes that things aren’t going well, and asks ChatGPT to change the rules, then regurgitate a new prompt capturing the alterations. It’s of glancing interest that even after the changes, the transcript remains awful: despite being told to stop prompting with an explicit list of options, ChatGPT keeps doing that; the stilted, buzzword-laden prose make it feel like you’re playing DnD with the worst, most corporate manager you’ve ever had; and there’s nothing resembling an actual conflict or revelation, just flaccid set-pieces and irrelevant revelations following each other in succession until the author declares that he’s won. To be fair, I guess I should note that I didn’t notice any glaring inconsistencies or logical contradictions in what the LLM spat out, which either indicates our forthcoming robot overlords are getting better with the hallucination problem or just that the “writing” was so soporific and arbitrary that there was no central narrative for individual developments to contradict.

But like I said, all those criticisms are only of glancing interest. I repeat: this prompt, which was submitted to the Comp as a thing you could use to get an LLM to play a game with you, is itself the product of an LLM – Penny Nichols, Troubleshooter is a coprophagous ouroboros, creating the very slop it feeds on, of no possible use or value to a human being. In that sense I suppose there’s something potentially meta to the prompt’s “final twist”: “Dragons feed on human attention, and this hoard has been feasting.”

They do, and it is.

Are we inclined to do something about that?

This is not IF.