Hey! Why would anyone read something that nobody actually wrote?!
That question comes up often when people talk about artificial intelligence. When a language model produces a company memo or a text message, you must acknowledge that no self-aware person wrote it. That’s true no matter how much prompting was involved. Nobody is writing these things. In the case of the memo, very little prompting went into it, either, because nobody cares.
So, what about the chatbots? For me, I have to think about my chatbots so to speak. I find these chatbots mildly interesting. Artificial voices lack the push-back of human essence, I'd agree. Yet I’ll read AI-generated character dialogue with the same half-curious attention someone might give a cheap novel.
SillyTavern, I mean. It’s silly (hence?), but it gives the same interest that a mobile game would. (In some installations, it is a mobile game...) However, if something looks like it was churned out by a marketing team or venture-backed startup, I lose interest immediately. That's true whether it was AI that wrote it, or the marketing team itself. I'm sure you'd react similarly.
Prompting itself takes a kind of narrative instinct. It’s similar, in a loose way, to how roleplay communities build shared stories. The process is smaller in scope, though. The presence of the robots, and the sheer complexity of it, makes it difficult to consider the process art in the same way that I consider roleplaying to be, though. Weird viewpoint, huh? Sorry. Maybe I'm wrong...
It’s incredibly important to remember that nobody really writes these messages. Still, they’re interesting to read sometimes, and the process keeps me engaged with computer science. Maybe that’s why the program is called SillyTavern? It’s a bit of a diversion. Many other distractions exist, and most don’t involve Markov chains. Technology keeps shifting, though, and I’m curious about it.
I often read these messages precisely because no one wrote them. The absence of an author makes them intriguing. Often the results are dull or messy. Sometimes they’re overly shaped by marketing despite the bot itself being the only product. Hmm. I guess that’s worth studying too. With careful prompting, you can produce text that vaguely implies a conversation.
It’s only worth reading, though, if you remember the nature of LLMs, and that nobody actually wrote the message. It can still be oddly charming. Maybe that’s the point, in an eerie sort of way…