Go to the game's main page

Review

1 of 1 people found the following review helpful:
Asylum escape game with LLM integration., September 18, 2026
Related reviews: about 2 hours

Doppelgangers

Man, this was kind of a bummer.

I've been pretty militantly anti-AI text adventures for a long time. I just think they're not that good, and lead to poorly made games. I do think some games I have enjoyed in this competition and ones earlier in this year contain text or plots that are AI-generated and human curated, and I believe some games I found mediocre have almost fully AI-written text. The writing has improved somewhat. But I still find most AI-designed games to be fairly poor.

At the same time, I have started using AI more in other areas of my life. I do like its ability to scan images and produce text from them, for instance a png of a manga panel and translating it into japanese, or asking it how to do a math problem not to trust its answer, but to see what method it reaches for and see if it's one I'm familiar with.

So I decided to give the LLM in this game a chance (you can choose to activate it or not to satisfy comp rules). I activated it. You have to sign in with an email. I don't trust AI sites with my email info so I made a burner gmail and used a rarely-used browser to log in.

Unfortunately, the LLM contributed essentially nothing, from what I saw of Mike Russo's playthrough.

The idea of this game is that you're wandering an asylum, while you have dementia, and you have to interact with others. You can talk to others to persuade them to do things. Occasionally, the game gives you persuasion checks. In non-AI mode, you always fail.

In AI mode, I guess you always fail. I tried many variations, including cajoling, threatening, reasoning, pointing out i was in a fictitious world. I couldn't find any way at all to pass the checks.

I did beat the game by being evil and killing everyone. But the LLM essentially contributed nothing; I don't see any reason to play with it on. Even with it on, it didn't print any AI-generated text, which was nice; instead, it chose between pre-written options based on the AI's decision.

The base game was janky; I dropped all in my room and got electroshock therapy; when I returned later, I couldn't see any items, but could TAKE ALL. Basic actions like taking your clothes are described in clunky ways. Many commands starting with TALK TO give the result of the game has the game say you can only go N, S, E, or W. There's a random maze at the end without much to redeem it.

I think the game could have benefitted a lot from seeing what other games can do in this space. Even if we kept the generic, stereotyped asylum setting, we had text games like Lunatix - The Insanity Circle from almost 30 years which had a rich custom parser, compelling characters and a fun plot.

The best redeeming feature of Doppelgangers is its interesting take on conversation and NPC independence, especially things like talking about other patients and things and the way that NPCs find bodies you don't hide well or take contraband into account. A game that combined these features with the polish and storyline of more mainstream IF games (if there is a mainstream for IF parser games) would be very solid.

By the way, I have been harsher in this review, and that's because I assume much of the coding and/or planning was AI generated, so is not due to the author's own inability, and can be more fairly pointed out.

Edit: I found out you can see the game code in the download, which is fascinating and fun to read, but is also clearly ai-influenced or written, unless the author puts emojis in the comments for fun. I do wonder if there is a bug with the LLM part, since I saw the prompt that gets sent to them and even following the prompt very closely I couldn't succeed in persuasion after logging in and using LLM mode.

It does hurt the game having such open-ended input with very little responsiveness to mistakes besides 'you failed'. You can't tell if you're failing because you need to alter your approach, or if the llm is broken, or if you need to capitalize letters, etc. That's one reason I don't really like LLMs in games in general, the hard boundaries of the world are often the most helpful, and tearing those down doesn't really enhance gameplay.

You can log in to rate this review, mute this user, or add a comment.