Are LLMs becoming similarly creative? Evidence from three years of models
Researchers analyzed three years of Large Language Model (LLM) performance on open-ended tasks. They found a statistically significant decrease in model output diversity over time, suggesting that LLM outputs may be converging in creative substance across models. This trend could lead to homogenization and diminish human agency in co-creative work with AI.
Researchers analyzed three years of Large Language Model (LLM) performance on open-ended tasks. They found a statistically significant decrease in model output diversity over time, suggesting that LLM outputs may be converging in creative substance across models. This trend could lead to homogenization and diminish human agency in co-creative work with AI.
---
Why it matters: This matters because it implies that the creative potential of LLMs may be limited by their increasing similarity in output. Engineers and researchers working on AI systems for human-AI collaboration need to consider this trend and its implications for the role of humans in the creative process.
Source: https://arxiv.org/abs/2608.19437
This article was originally published at: https://arxiv.org/abs/2608.19437