"Uncensored" open LLMs show higher optimism than base models, study finds

A recent study compared uncensored open LLMs with their base counterparts. The analysis measured optimism levels in model outputs. Uncensored versions consistently scored higher on optimism metrics. The findings suggest that content filters affect model sentiment. Researchers used a standardized set of prompts to evaluate responses. Results indicate a measurable shift toward optimism when filters are removed. The study highlights trade‑offs between safety controls and output tone. Future work may explore balancing optimism with responsible AI behavior.