A new study has revealed that large language models like ChatGPT struggle to accurately summarize scientific papers, often prioritizing simplicity over precision. When tasked with condensing complex research for news briefs, the models tended to oversimplify key findings, misrepresent methodologies, and occasionally introduce factual inaccuracies. The research, conducted by a team of science journalists and AI experts, involved comparing AI-generated summaries of peer-reviewed scientific articles with expert-written versions. The results showed that while the AI outputs were fluent and readable, they frequently omitted critical details, misrepresented statistical significance, or generalized results in ways that could mislead readers. One major issue identified was the tendency of LLMs to “sacrifice accuracy for simplicity,” distilling nuanced scientific conclusions into broad, accessible statements that lack the precision required for informed understanding. For example, models often failed to distinguish between correlation and causation, downplayed uncertainty in results, or misrepresented the scope of a study’s conclusions. The study also found that ChatGPT and similar models struggled with technical terminology, sometimes replacing precise scientific language with vague or incorrect alternatives. In several cases, the AI incorrectly attributed findings to the wrong study or conflated results from different experiments. Researchers noted that while AI tools can assist with drafting initial content or generating ideas, they are not yet reliable for producing accurate scientific summaries without expert review. The findings underscore the need for caution when using generative AI in science communication, particularly in media contexts where public understanding depends on factual integrity. The team concluded that human oversight remains essential, especially when translating complex research into public-facing content. They recommend that journalists and science communicators use AI as a supplementary tool—never a replacement—for ensuring accuracy and context in reporting on scientific advances.