chat

You could argue it was the low quality of the LLM, but I've seen the stuff platforms like Character.ai and Replika put out, and it isn't much better. You could argue it was the smaller userbase, and that could definitely be part of it, but I think there's a crucial factor at play here:

Everything about how AI Dungeon was promoted and presented.

AI Dungeon's original creator never anthropomorphized it, never pitched it as a companion, never talked about creating AGI, never invited debates about its sentience. He described it as exactly what it was: a procedural generation algorithm used to facilitate a freeform roleplaying game. The very first thing the interface did was prompt you for a genre and a character archetype. It created a character, gave that character a name, and from that point on, every time it addressed the player in second person, it was in such a way that anyone would read it as referring to the fictional character within the game's world. Its appearance matched its function: it was a Roguelike that procedurally generated text instead of level layouts.

As the community and platform developed (before OpenAI inevitably enshittified it), it also did so in a way that reinforced to those involved that the AI was a tool. People shared custom scenarios, scripts, advice on how to steer it toward session-appropriate outputs. It was talked about as a tool and treated as a tool, and the way the platform incorporated interfaces for things like generating and sharing scenarios.

I'm not going to go so far as to say that the only possible problem with LLMs is how they're marketed and pushed into everything, but I do think this is a striking example of how capitalism expresses every new technology in its most harmful and malignant form.