Skip to content

Week 41, 2026

← Week 40, 2026 · All weeks in the News Archive

Highlights selected from everything kept between October 2, 2026 and October 9, 2026. Items are selected several times a day by an automated pipeline and the summaries are machine generated; follow the links to the original sources. See About for how selection works.

The week in brief

Read aloud from this page's text by an AI-generated voice, not a human narrator; the text is authoritative.

At UT Southwestern Medical Center, a multimodal artificial intelligence (AI) system called MAPLES (Multimodal Assessment Pipeline for Learning Encounter Scoring) graded Objective Structured Clinical Examinations (OSCEs) for 222 students in Fall 2025, scoring their notes, audio, and video [3]. Faculty reviewed only the low-scoring tail, where AI agreement with standardized patient evaluators ranged from 85.7 to 92.4% by modality. In 616 selected disagreements, physician adjudicators matched the AI on 76.0% of items [3]. Human scoring passes fell by 92.3% against a modeled manual comparison, which makes AI first-pass grading an operational question for assessment offices rather than a hypothetical one [3].

Clinical skills that students practice are being tested the same way. In a randomized trial, a large language model (LLM) agent took better structured ophthalmology histories than residents, but residents gained more once ocular signs were added [2]. In a retrospective study, clinicians preferred LLM-written discharge summaries for 7- to 21-day stays in 95% of encounters [5]. A meta-analysis compared automated real-time coaching with expert instruction on simulated surgical tasks [6]. The Hong Kong Academy of Medicine issued guidance on AI in postgraduate training [8], while one podcast asked whether AI-improved student work reflects actual learning [13].

Benchmarks deserve caution. Among 127,833 clinician queries to an institutional assistant, documentation and administration made up 36.2% and diagnosis only 3.7%, yet the median public benchmark contained no documentation requests [7]. That gap matters when weighing reported gains for models such as MedGemma [1]. It also matters as Utah moves toward letting an AI prescribe acne medication without prior clinician review, subject to third-party audits [4].

Written from the items highlighted below; numbered links go to the sources.

New opportunities

Color key: News and Events

Medical Education

npj Digital MedicineOctober 7, 2026
Prospective deployment of multimodal AI grading for medical student OSCEs

A prospective study reports deploying multimodal artificial intelligence to grade medical student objective structured clinical examinations OSCEs , evaluating its use in real assessment.

PubMed AI in medical educationOctober 8, 2026
Real-Time Intelligent Coaching Versus Human Expert Instruction in Surgical Skills Training: A Systematic Review and Meta-Analysis

A systematic review and meta-analysis compares real-time automated coaching systems with human expert instruction for surgical skills training in simulated tasks.

PubMed AI in medical educationOctober 7, 2026
Hong Kong Academy of Medicine position statement on artificial intelligence in postgraduate medical education

The Hong Kong Academy of Medicine issued a position statement on how artificial intelligence should be used in postgraduate medical education.

Clinical Practice

Nature MedicineOctober 6, 2026
An open vision-language model for diverse medical applications

Nature Medicine publishes MedGemma, an open collection of medical vision-language models based on Gemma 3 that outperform similarly sized generative models on medical image and text tasks.

npj Digital MedicineOctober 5, 2026
Standardized pre-consultation by a large language model agent vs ophthalmology residents: a randomized clinical trial

A randomized clinical trial compared a large language model agent with ophthalmology residents for standardized pre-consultation, assessing how well the AI collects patient history before the visit.

STAT News AIOctober 5, 2026
STAT+: Utah plows ahead with more health AI pilots for prescriptions, women’s health

Utah plans to let an AI product evaluate patients and write new drug prescriptions without prior human review, expanding its health AI pilot programs to include women's health.

npj Digital MedicineOctober 2, 2026
Clinician-centered evaluation of large language model-generated discharge summaries for longer hospitalizations

A peer-reviewed study has clinicians evaluate discharge summaries generated by large language models for longer hospital stays.

arXiv cs.CLOctober 9, 2026
Clinician use of language models diverges from how the models are evaluated

A preprint argues that how clinicians actually use large language models differs from the exam-style benchmarks used to judge their readiness, questioning whether benchmark scores predict real deployment performance.

Videos

Video: Stanford MedAI seminar on an agentic biomedical question answering system Stanford MedAI seminar on an agentic biomedical question answering system Stanford MedAI, October 7, 2026 Stanford MedAI seminar where Nhung Nguyen presents Med.ai ASK, an agentic system that answers biomedical questions, with discussion of its design and evaluation. Video: OpenAI introduces GPT-6 in ChatGPT with Intelligent UI OpenAI introduces GPT-6 in ChatGPT with Intelligent UI OpenAI, October 7, 2026 OpenAI announces GPT-6 in ChatGPT along with Intelligent UI, which lets ChatGPT answer with interactive interface elements rather than only text; the video demonstrates the new feature. Video: Overview of DeepMind's AlphaGenome Atlas for genome interpretation Overview of DeepMind's AlphaGenome Atlas for genome interpretation Two Minute Papers, October 7, 2026 Covers Google DeepMind's AlphaGenome Atlas, an AI system for interpreting genome sequences, with an overview of what it predicts and why it matters for genetics research. Video: Explainer on distributing large AI models across multiple GPUs Explainer on distributing large AI models across multiple GPUs IBM Technology, October 6, 2026 An IBM presenter explains why the largest AI models cannot fit on one graphics processing unit GPU and how work is distributed across multiple GPUs, giving a conceptual overview of scaling.

Podcasts

Podcast: The AI Learning Paradox: Better Results, Less Learning? The AI Learning Paradox: Better Results, Less Learning? AI in Education Podcast, October 8, 2026 Hosts Ray and Dan discuss whether AI helps students learn or only produces better work, covering Australian universities' AI announcements UNSW, Flinders, University of Sydney and recent research on learning outcomes. Podcast: A.I. Agents: Cute, Cuddly and Maybe Catastrophically Dangerous? A.I. Agents: Cute, Cuddly and Maybe Catastrophically Dangerous? Hard Fork, October 2, 2026 Hard Fork discusses the growing use of AI agents and the safety risks they may pose, offering general AI literacy on autonomous software tools. Podcast: The Best Way to Test New AI Models The Best Way to Test New AI Models The AI Daily Brief, October 7, 2026 Nufar Gaspar explains a repeatable method for testing new AI models on your own tasks, comparing output quality, speed, and cost to decide which models belong in your workflow.

Also this week

Assessment and feedback

Attitudes and adoption

Benchmarks and evaluation

Clinical decision support

Clinical documentation

Exams and benchmarks

Imaging and diagnostics

Industry and policy

New models

Patient facing tools

Research and methods

Safety and evaluation

Safety and reliability

Simulation and skills

Teaching and curriculum

Other

← Week 40, 2026 · All weeks in the News Archive

Report a problem with this page

Comments

Joining the discussion requires a free GitHub account. Comments are stored publicly on the site's GitHub Discussions board, so leave out anything you would not post in public, such as patient details, student records, or exam content. Comments are moderated.