

Even less plausible. There was a paper published recently arguing that by design LLMs are quite literally incapable of creativity. These predictive statistical models represent averages. They always and only generate the most banal outputs. That’s what makes them useful.






That’s a good point. The problem is that LLMs are calibrated for prediction. Their randomness is tweaked for efficacy. Forcing them to be more chaotic just makes them less effective. This inherent tension is why they’re mathematically incapable of consistent creativity.