United KingdomIn late 2024, artificial intelligence achieved PhD-level performance on scientific-reasoning benchmarks, and by 2025, it approached the measurement ceiling for one such test. The author, who directs AI initiatives at Vanderbilt University, developed an AI module to train biomedical PhD students in prompt engineering. A survey revealed that while 81% of students used AI tools for science, only 5% were proficient in writing effective prompts, rising to 48% after the module. The author emphasizes that scientific training can enhance AI usage without requiring computational expertise. They advocate treating each AI prompt as an experiment, offering strategies such as connecting tools via the Model Context Protocol, structuring inputs with Markdown/XML/JSON, and using meta-prompting to let the AI generate optimal prompts.
Bias read (Center): The article discusses AI development and educational practices in a scientific context, focusing on technical methodologies rather than ideological positions. While it touches on academic institutions and research, there is no overt political framing or advocacy for specific policies. The emphasis在于
Why factuality (60): The article discusses AI usage in academic settings and personal teaching experiences but does not reference the GPQA dataset or the specific results mentioned in the primary source document. It lacks direct alignment with the benchmark data provided.
Why objectivity (75): The tone is positive and encouraging towards AI integration in education, focusing on practical strategies. There is no overt bias or emotional language, though the focus on personal experience may slightly skew perspective.
