alt Hacker NewsA quarter of a century ago we used to do this on IRC, by tuning markov chains we'd fed with stuff like the Bible, crude erotic short stories, legal and scientific texts, and whatnot. Then have them chat with each other.
A quarter of a century ago we used to do this on IRC, by tuning markov chains we'd fed with stuff like the Bible, crude erotic short stories, legal and scientific texts, and whatnot. Then have them chat with each other.
At least in my grad program we called them either "textural models" or "language models" (I suppose "large" was appended a couple of generations later to distinguish them from what we were doing). We were still mostly thinking of synthesis just as a component of analysis ("did Shakespeare write this passage?" kind of stuff), but I remember there was a really good text synthesizer trained on Immanuel Kant that most philosophy professors wouldn't catch until they were like 5 paragraphs in.