AI systems that forget everything after every conversation are actually more creative than models that remember who you are.

Static, memoryless AI systems are significantly more likely to generate innovative outcomes than adaptive, memory-based ones. We usually assume that an AI learning from previous interactions is a good thing, but this adaptation actually leads to more boring, predictable results. Memory-based systems tend to converge on safe, familiar answers based on past feedback. A blank slate model is free to explore wilder, more unconventional ideas that a memory-based system would filter out. This research suggests that dumb AI might be a better partner for high-level creative work than smart adaptive assistants.