Students can manipulate #LLMAsAJudge to change #Assessment output. #PromptInjection changed 29% LLM scores & 98% narrative #feedback. Major problem for #AISecurity #LLMSecurity #GenAISecurity #AiinEducation #TrustworthyAI #Jailbreaking #AdversarialAI medrxiv.org/content/10.648…
David A. Cook
@CookMedEd
Medical education researcher, internist @MayoClinic. #meded Assessment, Online Learning, Education Technology, Simulation, Research Design. @LDSchurch
10 Following 985 Followers 61 Posts
Latest posts 5
Can accurate LLM scoring still be unsafe? Students can control grading thru prompt injection. This #AIRedTeaming study shows #PromptInjection changes output across 4 LLMs. #MedEd #AISafety #AIRisk #Cybersecurity #Validity medrxiv.org/content/10.648…
Trying to write a better prompt for #ChatGPT #Gemini #Claude or another #LLM? New arXiv paper shares evidence-based principles of #PromptEngineering. Novel #PICCO prompt framework + review of techniques & security, privacy, bias, trust. Read it: arxiv.org/abs/2604.14197 #MedEd #AI
Can #AI help with #qualitative data analysis? Absolutely! but it's complicated. This article discusses promises, pitfalls, and a variety of approaches. pubmed.ncbi.nlm.nih.gov/40560241/ #CAQDAS #QDA #MedEduc #LLMs
How do physicians develop "wisdom" (#phronesis)? This qualitative study explored how surgeons develop ability to combine knowledge + technical skill to make wise clinical decisions. #MedEduc #medicaldecisionmaking #clinicalreasoning authors.elsevier.com/c/1lJme90L-3I9X