Study Finds AI Language Models Enhance Novice Performance in Complex Biology Tasks
Research indicates that large language models significantly improve the accuracy of novices tackling biosecurity-related biology problems, raising considerations for scientific progress and safety.
Recent research indicates that large language models (LLMs) significantly enhance the performance of individuals with limited biology experience when tackling complex, biosecurity-related tasks. In a study involving participants working on eight challenging biology problems, those with access to LLMs were over four times more accurate than those relying solely on internet resources. Notably, in three out of four benchmarks where expert performance data was available, LLM-assisted novices matched or exceeded expert accuracy. However, standalone LLMs often outperformed human-LLM collaborations, suggesting that users may not be fully leveraging the models' capabilities. Additionally, 89.6% of participants reported minimal difficulty in accessing dual-use information, despite existing safeguards. These findings underscore the need for ongoing evaluations of LLMs' impact on both scientific advancement and biosecurity.
