In class today we mentioned how LLMs may preferentially use B and C as the MCQ answers. So why is that the case? Is the distribution the same for different LLMs? How is it influenced by the prompt, model parameter size, architecture, and even the company that trains the model?
We can further extend this idea to other choices. For example, if asked to generate a random number, what number will LLMs choose? Will it be more prone to generate integers or floats, small or larger numbers, positive or negative numbers?
If you are inspired by this idea, you can reach out to the authors for collaboration or cite it:
@misc{chen-how-llms-make-2026,
author = {Chen, Mike},
title = {How LLMs make choices},
year = {2026},
url = {https://hypogenic.ai/ideahub/idea/Qo10WXGm3ANrycOq2e1p}
}Please sign in to comment on this idea.
No comments yet. Be the first to share your thoughts!