I keep brooding on the idea that maybe LLMs will never get “good at writing” in some Platonic sense, but they will be progressively trained better to create text that a greater and greater bulk of humans will “enjoy” for reasons that are downstream of inherited patterns of thought and emotional response that the models have learned.
I keep brooding on the idea that maybe LLMs will never get “good at writing” in some Platonic sense, but they will be progressively trained better to create text that a greater and greater bulk of humans will “enjoy” for reasons that are downstream of inherited patterns of thought and emotional response that the models have learned.
In other words, we become the stochastic parrots.