LLM-s are fascinating tech, but there needs to be another development with them that makes them factual. As they are now, they're good for fiction only, and even then you need to curate their output.
Curiously, when I tried out Deepseek (I think the 14b parameter model) I had a conversation about Peter Watt's Blindsight and the nature of conscience, intelligence and AI. It distilled and represented some good points various people have made about these concepts (because that's what LLM-s do). Then I asked it to recommend me other such works in literature, film and theater, and its list was pretty solid. Except one theatrical play it recommended to me. The playwright is real, the theater where it was first performed is real, when I asked Deepseek more about the play—called "A machine that generates desire"—the characters and themes were all compelling, the concept of desiring-production is real. The only problem: as much as I searched, a play with that title does not exist. But I would actually like to go and see that play
Curiously, when I tried out Deepseek (I think the 14b parameter model) I had a conversation about Peter Watt's Blindsight and the nature of conscience, intelligence and AI. It distilled and represented some good points various people have made about these concepts (because that's what LLM-s do). Then I asked it to recommend me other such works in literature, film and theater, and its list was pretty solid. Except one theatrical play it recommended to me. The playwright is real, the theater where it was first performed is real, when I asked Deepseek more about the play—called "A machine that generates desire"—the characters and themes were all compelling, the concept of desiring-production is real. The only problem: as much as I searched, a play with that title does not exist. But I would actually like to go and see that play
