In partnership with
Join 400,000+ executives and professionals who trust The AI Report for daily, practical AI updates.
Built for business—not engineers—this newsletter delivers expert prompts, real-world use cases, and decision-ready insights.
No hype. No jargon. Just results.
Hi!
Today is LLM Friday.
A day when I will only share Research, Tools, and News on Large Language Models in Medicine.
A few days ago, Nature published an article about Delphi-2M, an AI predictive model trained with data from 400,000 participants from the UK.
The model uses health records and lifestyle factors to estimate the likelihood to develop more than 1000 diseases. It can simulate possible health trajectories for up to 20 years.
The model was validated with 1.9 million Danish patients, but they will keep testing and validating the model with datasets from other countries.
Research like this helps advance predictive medicine.
Are you ready to dive into today’s issue?
✨LLMs
Researchers tested a conversational AI system that interviews patients and creates medical notes for physicians to review and approve.
They wanted to see if this setup could maintain safety while reducing doctor workload.
🔬 Methods
The study had 3 steps:
Interviews
Oversight by experienced blinded physicians
Post evaluation by another group of blinded physicians
Interviewers:
AMIE: A Multi-agent AI system.
Control group 1: 20 licensed physicians across multiple specialties with less than 5 years of practice.
Control group 2: 10 Nurse practitioners and 10 Physician assistants.
60 clinical cases.
The 3 groups interviewed patient actors via text to obtain (without giving clinical advice):
Clinical history
Drafted SOAP notes
Proposed a diagnosis
Management plan
Oversight:
20 experienced licensed physicians with a median of 12 years of experience who edited notes and sent messages to the patients or requested follow-up consultations.
Blind physicians:
19 blinded physicians rated the final results.
📊 Results
g-AMIE produced more complete histories and summaries and better proposed diagnoses and treament plans.
Supervision of the AI system was 40% faster than traditional
consultations.
AI notes were much longer, creating mixed results considering the physician burden.
The AI system avoided giving advice in 90% of cases.
Communication & empathy were rated higher for g-AMIE in comparison to controls.
It’s important to consider that the clinicians give advice in a normal clinical setting, or don’t need to complete certain checkboxes for differencial diagnosis.
These are all steps that the AI system was instructed to follow.
🔑 Key Takeaways
This asynchronous model showed how this AI system has the potential to handle initial consultations.
Physician oversight is critical for the safe integration, accuracy, and context.
The system can save time and support physicians, but it needs to be validated in real-world clinical settings.
💡AI should augment, not replace, human expertise.
🔗 Vedadi E, Barrett D, Harris N, Wulczyn E. Towards physician-centered oversight of conversational diagnostic AI.arXiv. Posted July 24, 2025. https://doi.org/10.48550/arXiv.2507.15743
🦾TechTools
Evidence-based search across Elsevier’s textbooks and Journals.
Can create summaries, has updated guidelines and procedure videos.
Has plans depending on the medical job and specialty.
Can be integrated into the Electronic Health Record.
Available by subscription (offers a free trial)
An AI Medical Diagnosis Assistant. You can upload medical images and describe symptoms to get diagnosis assistance.
Has been tested in benchmarks for radiology, dermatology, ophthalmology, and pathology.
Supports multiple languages. HIPAA compliant.
Free for low use, also has subscription plans.
Draft differential diagnoses and treatment plans using evidence-based guidelines.
Update medical guideline databases regularly to ensure reliability.
Summarizes notes, history, medications, laboratory and imaging data.
Integrates with electronic health records.
For institutional use.
That’s all for today.
You’re already ahead of the curve in medical LLMs — don’t keep it to yourself. Forward AIMedily to a colleague who’d appreciate the insights (Thank you!)
Enjoy the weekend!
Until next Wednesday.
Itzel Fer, MD PM&R
Forwarded this email? Sign up here
P.S. Is there something you would like me to cover in the newsletter? Hit reply with your answer (in any language). I promise I will read it.
How did you like today's newsletter?
Fact-based news without bias awaits. Make 1440 your choice today.
Overwhelmed by biased news? Cut through the clutter and get straight facts with your daily 1440 digest. From politics to sports, join millions who start their day informed.
