A new study published on arXiv has found that large language models exhibit significant concentration in their outputs when asked to generate random lottery numbers. Across 1,200 attempts using six different language model configurations, the models produced a limited diversity of number combinations, with modal tickets accounting for a substantial percentage of valid responses. This concentration is notably higher than what would be expected from independent, uniform random sampling, suggesting a bias in how these models generate seemingly random sequences. AI
IMPACT Suggests potential biases in LLM's random number generation, impacting applications requiring true randomness.
RANK_REASON The cluster contains a research paper published on arXiv detailing experimental findings. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
