AI Can’t Help Real People - But Pass Med Exam w 95%

We need to code new AI systems w/ improved Human-AI interaction, especially on medical topics and health care. We need to create new training datasets for pre-training new VLMs for clinical settings and improve medical reasoning abilities (w/ dedicated data training sets) for incomplete human information transfer in medical domains. New question to solve for insurance companies: What company is responsible regarding AI liabilities and high insurance costs: the hospital or Anthropic (w/ their "Ai model welfare" when AI models become self-conscious)? Are Ai models liable themself (nice idea, Anthropic - smile) or their creators? Your thoughts? LLMs ace benchmark tests but flunk real-world patient interactions. All rights w/ authors: "Clinical knowledge in LLMs does not translate to human interactions" by Andrew M. Bean1, Rebecca Payne2,3, Guy Parsons1, Hannah Rose Kirk1, Juan Ciro4, Rafael Mosquera5,6, Sara Hincapi´e Monsalve5,6, Aruna S. Ekanayaka7, Lionel Tarassenko8, Luc Rocher1, Adam Mahdi1 from 1 Oxford Internet Institute, University of Oxford, Oxford, UK. 2 Nuffield Department of Primary Care,University of Oxford, Oxford, UK. 3 North Wales Medical School, Bangor University, Bangor, UK. 4 Contextual AI, Mountain View, USA. 5 MLCommons, San Francisco, USA. 6 Factored AI, Palo Alto, USA. 7 Birmingham Women’s and Children’s NHS Foundation Trust, Birmingham, UK. 8 Institute of Biomedical Engineering, University of Oxford, Oxford, UK. #healthcareinnovation #healthcareprofessionals #airesearch #clinic #medicalcoding #logicalthinking