{
  "abstract": "We read with great interest the study by Carlini et al on using multimodal large language models (MLLMs), specifically GPT-4o and Gemini 1.5 Pro, to detect colorectal polyps using the SUN database.1 The authors present a provocative proof of concept: models trained on general internet-scale data can achieve high specificity and, in the case of GPT-4o, sensitivity approaching that of regulatory-approved CADe systems. However, while these findings highlight the ‘emergent’ medical intelligence of MLLMs, several statistical and clinical hurdles must be addressed before such models can be considered for integration into endoscopic workflows.2",
  "authors": [
    {
      "affiliations": [
        "Pediatrics, Ben-Gurion University of the Negev, Beer-Sheva, Israel"
      ],
      "name": "Zvi Weizman"
    }
  ],
  "title": "Large language models for detecting colorectal polyps in endoscopic images—further research is required",
  "uid": "599374f9-af89-540a-a423-8c59bb8bbb96"
}