P.110 Evaluating AI performance in written neurosurgery exams: a comparative analysis of large language models
Bibliographic record
Abstract
Background: The integration of Artificial Intelligence (AI) in medical education is an area of growing importance. While AI models have been evaluated extensively in multiple-choice question formats, their proficiency in written exams remains to be explored. Methods: Four AI models—GPT-4 (OpenAI), Claude-2.1 (Anthropic), Gemini Pro (Google), and Perplexity 70B (Perplexity)—were tested using the Canadian Royal College Sample Neurosurgery Exam. The written exam covered diagnostic reasoning, knowledge of neurosurgical conditions, and understanding of radiographic imaging techniques. Results: GPT-4 and Perplexity 70B both achieved a score of 68.42%, followed by Claude-2.1 with 60.53%, and Gemini Pro with 57.89%. The models showed proficiency in answering questions that required factual knowledge, such as identifying pathogens in spinal epidural abscess. However, they struggled with more complex diagnostic reasoning tasks, particularly in explaining the pathophysiology behind a sudden rise in blood pressure during surgery and interpreting radiographic characteristics of intracranial abscesses on MRI. Conclusions: The findings indicate that while AI models like GPT-4 and Perplexity 70B are adept at handling factual neurosurgical questions, their performance in complex diagnostic reasoning in a written format is less consistent. This underscores the need for more advanced and specialized AI training, particularly in the nuances of medical diagnostics and decision-making.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.023 | 0.109 |
| Meta-epidemiology (narrow) | 0.002 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.003 |
| Bibliometrics | 0.004 | 0.003 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.006 | 0.005 |
| Open science | 0.002 | 0.003 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.005 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".