ChatGPT Repetition Fix: Boost Creative AI Answers

Generative AI models are producing answers that are predictable and less diverse. This is a trade-off between quality and diversity, as efforts to make AI richer and safer can lead to less creative outputs. For example, education students at Western Galilee College asked AI systems to suggest additional characteristics of gifted students. Many responses were strikingly similar, even from different language models like ChatGPT and Gemini. This pattern appeared in other questions requiring creative thinking. Frequent users of generative AI notice this lack of variety, a phenomenon researchers call "mode collapse." Early text models produced more varied answers but made more mistakes. Now, reinforcement learning from human feedback, or RLHF, aims to improve results by ranking outputs based on human preferences. However, this process can lead to the AI generating more consistent, but less diverse, responses. This matters because it impacts the creative potential of AI tools.