A detective game makes lying LLMs the attraction

The sharpest objection was not that the suspects might lie, but that dealing with them already felt like a job. Two commenters took that position: loloquwowndueo said “Talking to lying LLMs is what I do all day for work.” and hxugufjfjf agreed, adding screen fatigue to the complaint. For these two, the central mechanic was the reason to stay away rather than the attraction.

In response to the complaint about screens, mikepurvis proposed a different setting: “I wonder if you could do it all over voice, say while alone in the car.” The site already advertises typed or spoken questions, but an entirely voice-driven session was the possibility he raised. His reply shifted the question from whether to interrogate a model to how that interrogation might fit into leisure.

Alongside the two fatigue objections, three commenters found something worth pursuing through older games or different interfaces. The nostalgia began with reactordev, who recalled Sherlock Holmes Consulting Detective on an early Macintosh and wanted that kind of crime thriller to work ad hoc. In reply, gregsadetsky supplied an archive link to the game. Separately, ghgr welcomed the concept through another reference point: “It reminds me of the 1985 videogame Where in the World Is Carmen Sandiego.”

Answering reactordev’s wish for flexible mysteries, mikepurvis drew a narrower target than unrestricted generation: “People are definitely doing "free form" interactive fiction now, though I think the one that really interests me is classic style but with an LLM interface that makes the interaction less rigid, eg” He linked an interactive-fiction discussion about LLM-based input parsing. That distinction put the interface, rather than an endlessly improvised plot, at the center of the appeal—and made the quality of the words on screen matter more.

The most specific criticism concerned the sales pitch rather than the game’s architecture. Because a reading-heavy detective game needs convincing dialogue, unintelligibled asked for stronger examples in the first trailer: “You should probably put some better examples of writing in the first trailer, especially if the game involves a lot of reading.” The commenter also objected that both displayed dialogue sentences used the same word, honestly. This was one reader’s criticism, but it targeted the material players would have to examine closely to catch a lie.

The founder’s own emphasis was on playability: a model plays detective and must solve the case before publication. The site describes a loop of searching places, questioning suspects and holding evidence against statements before submitting a report. Those are the project’s stated checks and mechanics; the trailer critique asked for a different assurance, visible in the writing itself.

The comments establish two distinct reactions, not a consensus. Two people associated the premise with work and screen fatigue; two others reached for fondly remembered detective games. A third interested participant, mikepurvis, explored voice play and a less rigid interface.

The attraction remained disputed because the same mechanic carried opposite associations: tiresome model conversations for some, more flexible investigation for others. The separate trailer criticism left a concrete question about whether the dialogue could sustain that investigation.

The founder’s next stated milestone was Apple’s review of the iOS version; Android was already free at launch. Weekly cases and private group games were longer-term ambitions conditional on the project taking off. For the concerns raised here, the revealing material would be the cases themselves: their dialogue, clues and answers.