Skip to content
Roadmap for Healthcare AI
Track 1

AI-literate clinician

2–4 weeks, no code

Test your knowledge15 questions on this track, with explanations

Read these five first

Landmark papers for this track, in reading order, from the papers list. Read each one critically with the 15-question checklist.

  1. Large language models in medicine #171
    how LLMs work
  2. Benefits, Limits, and Risks of GPT-4 as an AI Chatbot for Medicine #172
    real GPT-4 errors, so verify everything
  3. Accuracy and Reliability of Chatbot Responses to Physician Questions #174
    free; confident wrong answers
  4. Large Language Model Influence on Diagnostic Reasoning: A Randomized Clinical Trial #178
    giving doctors an LLM ≠ better diagnosis
  5. Comparing Physician and Artificial Intelligence Chatbot Responses to Patient Questions Posted to a Public Social Media Forum #173
    patient messaging and its limits

Start here, free

The free-first path for this track from the free-resources addendum, in order.

  1. Prompt engineering overview #98
  2. Prompting best practices #99
  3. Interactive Prompt Engineering Tutorial (9 chapters + appendix) #100
  4. Anthropic Academy (free courses incl. AI Fluency, Claude 101) #101

Topics explained

Plain-language explanations for each topic. Examples are illustrative, not real cases.

1

Prompting fundamentals

In plain words

A prompt is the instruction you give an AI chatbot. Clear prompts get better answers. Say who the answer is for, what you want, what to include or leave out, and what format you want back. If the first answer misses, refine the prompt rather than starting over.

Healthcare example (illustrative)

Weak: "Explain spinal anaesthesia." Stronger: "Explain spinal anaesthesia to a 60-year-old patient having a knee replacement, in under 150 words, at a reading age of about 12. Cover what they will feel, how long the numbness lasts in general terms, and when to call a nurse. Do not give drug doses."

A little more technical

Chatbots continue text based on patterns. The more context you supply (audience, purpose, constraints, an example of the output you want), the less the model has to guess. Useful habits: give one or two examples, ask for a fixed structure (headings, a table, bullet points), and split a big task into smaller steps.

Don't learn this yet

Skip "prompt libraries", lists of magic phrases, and advanced techniques like prompt chaining for agents. Clear instructions do most of the work.

2

healthcare prompting patterns

In plain words

A few prompt patterns recur in clinical work: summarising, explaining for a specific audience, drafting a letter or leaflet, comparing options, and critiquing a draft. Each works better when you supply the source material and ask the AI to stay within it.

Healthcare example (illustrative)

  • Summarise from a source: paste a guideline section and ask, "Summarise only what this text says about post-operative nausea, and quote the sentence behind each point."
  • Explain for an audience: "Rewrite this consent paragraph for a patient with limited English, keeping every risk mentioned."
  • Critique: "Here is my teaching slide on fluid management. List anything unclear or possibly outdated, and say why."

A little more technical

Grounding the model in text you provide ("answer only from the text below") reduces invented content, because the model can copy from the source instead of recalling from memory. Asking it to quote the exact supporting sentence makes checking quicker. Asking it to say "not stated in the text" when something is missing gives it a way to admit gaps.

Don't learn this yet

Leave building your own guideline chatbot (retrieval-augmented generation, or RAG) until Track 6.

3

hallucination and verification

In plain words

AI chatbots sometimes produce statements that sound confident and are wrong. This is called hallucination. It includes invented references, wrong doses and non-existent guidelines. Treat every factual claim as unverified until you have checked it against a trusted source.

Healthcare example (illustrative)

You ask a chatbot for evidence supporting a drug choice. It returns a neat citation with authors, journal and year. When you search for it, the paper does not exist, or it exists but says something different. The answer looked equally confident either way.

A little more technical

Language models generate the most likely next words; they do not look facts up unless they are connected to a search or document tool. So fluency is not evidence of accuracy. Practical checks: ask for sources and open every one; ground the model in documents you supply; ask it to separate what the source says from its own inference; and double-check numbers, doses and names in particular.

Don't learn this yet

You don't need to know why models hallucinate at a technical level. The habit that matters is "verify before use". Formal evaluation methods come in Track 3.

4

PHI rules for consumer tools

In plain words

Protected health information (PHI) is anything that could identify a patient together with their health details. Consumer chatbots and apps may store what you type, and may use it to improve their products. Unless your organisation has approved the tool for patient data, do not enter identifiable patient information.

Healthcare example (illustrative)

Pasting a clinic letter with the patient's name, date of birth and hospital number into a free chatbot to "tidy it up" is a data-protection risk. A safer route is to remove identifiers first and write a generic version ("a 54-year-old man with…"), or to use a tool your hospital has approved for patient data.

A little more technical

Removing names is often not enough: rare conditions, exact dates, locations and free-text details can re-identify someone. What is allowed depends on the law in your country, your employer's policies and the contract your organisation has with the tool's provider. In India, personal data is governed by the Digital Personal Data Protection Act 2023 (see #40). When in doubt, ask your information-governance or data-protection lead before using a tool with patient data.

Don't learn this yet

You don't need to study data-protection law in detail. Learn your own organisation's policy and the one rule: no identifiable patient data in unapproved tools.

5

AI tool map (clinical / research / education / productivity)

In plain words

AI tools fall into four groups by what you use them for. Clinical: literature search, documentation help, patient-education drafts. Research: searching and screening papers, extracting data, help with analysis and writing. Education: making questions, cases and teaching material. Productivity: email, slides, spreadsheets and meeting notes. The same general chatbot can often do several of these; specialist tools do one job with more safeguards.

Healthcare example (illustrative)

For one teaching session you might use a general chatbot to draft ten practice questions (education), a literature-search tool to find recent reviews (research), and a slide tool to lay out the talk (productivity). You would check every question's answer yourself before using it.

A little more technical

When comparing tools, ask: where does my data go? Does it show sources for its claims? Is it approved by my organisation? What does it cost, and what is limited on the free tier? Is it intended for clinical use, or explicitly not? A dedicated Tools section with verified details for individual tools comes in a later phase.

Don't learn this yet

Don't try to learn every new tool. Get good at one general assistant and one literature tool, and learn how to judge any new tool against the questions above.

Go deeper (optional)

6

basic workflow automation (no-code)

In plain words

Workflow automation connects steps you currently do by hand, so that one event triggers the next without you copying and pasting. "No-code" tools let you build these chains by clicking rather than programming. AI can be one step in the chain, for example summarising or sorting text.

Healthcare example (illustrative)

A department's teaching coordinator gets session feedback through an online form. An automation could collect each new response, ask an AI step to group the free-text comments into themes, and add the result to a shared spreadsheet for the tutor to review. No patient data is involved, and a person reviews the output before acting on it.

A little more technical

Automations are built from a trigger (a new form response, email or file), actions (copy, transform, send) and, where needed, a human approval step. With AI in the chain, add a checkpoint where a person reviews the AI's output before anything is sent or saved permanently. Keep patient data out unless the whole chain is approved for it.

Don't learn this yet

Leave APIs, webhooks and multi-step AI agents until Track 6. Start with one small, low-risk, non-patient task.

Resources for this track

Every resource here has had its link and relevance checked, except those tagged “Link not machine-checked”. “Not listed” means the detail wasn't stated or couldn't be confirmed.

The resource list groups Tracks 0 and 1 together, so these resources appear on both.

Type
Depth
Coding
Focus
Era
Cost
Region

Entries that don't state a value only appear when that filter is set to “Any”. “Free only” shows entries whose cost is stated as free, plus everything from the free-resources addendum.

Showing 30 of 30
#1
Generative AI for Healthcare
Google (Google Skills; free access via DiMe)
UnderstandUseLLM-eraHealthcare-specific

AI/ML/DL basics, gen AI and LLMs in healthcare, prompt design for healthcare Q&A, Google Cloud tools (Vertex AI Agent Builder, Model Garden). LLM+. Note: Google says Vertex AI is now "Gemini Enterprise Agent Platform", so tool names are changing.

Cost
Free via DiMe (survey first)
Duration
Not listed
Level
Coursera lists Intermediate
Cert / CME
Skill badge; Coursera version has a shareable certificate
More details
Format
Self-paced + hands-on lab
Code
N
HC
Y
Depth
U, Us

Resource-list status: V
Listed under: Track 0–1: Orientation and AI-literate clinician (prompting, tools, PHI, no-code)

Link checked: 2026-09-27 · Details researched: 26 Sep 2026Report a problem
#2
Google AI Essentials
Google
UnderstandUseLLM-eraGeneral AI

5 modules: intro to AI, productivity tools, prompting, responsible AI, staying current. LLM+ (2024).

Cost
India: US$20/month after 7-day free trial (US/Canada US$49/month)
Duration
"Under 10 hours" (India page)
Level
Beginner
Cert / CME
Google certificate
More details
Format
Self-paced
Code
N
HC
G
Depth
U, Us

Resource-list status: V
Listed under: Track 0–1: Orientation and AI-literate clinician (prompting, tools, PHI, no-code)

Link checked: 2026-09-27 · Details researched: 26 Sep 2026Report a problem
#3
YUVA AI for ALL
IndiaAI Mission, MeitY (Govt of India)
UnderstandUseLLM-eraGeneral AIIndia

Six modules: what AI is, the technology behind it, the CRAFT prompting formula, gen-AI tools, ethics, future. Indian examples. Launched Nov 2025; Coursera page updated Jan 2026. LLM+.

Official URL
https://www.futureskillsprime.in/course/yuva-ai-for-all/ (also Coursera, iGOT Karmayogi, YouTube)
Cost
Free
Duration
Not listed
Level
Beginner
Cert / CME
Government of India certificate (FutureSkills Prime); Coursera certificate
More details
Format
Self-paced
Code
N
HC
G
Depth
U, Us

Resource-list status: V
Listed under: Track 0–1: Orientation and AI-literate clinician (prompting, tools, PHI, no-code)

Link checked: 2026-09-27 · Details researched: 26 Sep 2026Report a problem
#4
AI Foundations for Everyone (Specialization)
IBM / Coursera
UnderstandLLM-eraGeneral AI

AI/ML/DL basics, gen AI and prompting, IBM watsonx, no-code AI. LLM+.

Cost
Coursera subscription; free enrolment/audit; in Coursera Plus
Duration
4 weeks at 10 h/week (4 courses)
Level
Beginner
Cert / CME
IBM certificate
More details
Format
Self-paced
Code
N
HC
G
Depth
U

Resource-list status: V
Listed under: Track 0–1: Orientation and AI-literate clinician (prompting, tools, PHI, no-code)

Link checked: 2026-09-27 · Details researched: 26 Sep 2026Report a problem
#5Link not machine-checked
AMA ChangeMedEd® Artificial Intelligence in Health Care Series (7 modules)
AMA + University of Michigan DATA-MD
UnderstandClassical (pre-LLM)Healthcare-specific

Intro to AI/ML and big data; methodologies; practical applications in the health system; AI and precision health; ethics/legal. Aimed at medical students, also for residents and physicians. Mostly Classical.

Cost
Free
Duration
Not listed
Level
Beginner
Cert / CME
Not listed
More details
Format
Self-paced microlearning
Code
N
HC
Y
Depth
U

Resource-list status: V
Listed under: Track 0–1: Orientation and AI-literate clinician (prompting, tools, PHI, no-code)

Link not machine-checked · Details researched: 26 Sep 2026Report a problem
#6Link not machine-checked
Artificial and Augmented Intelligence in Health Care (course 318)
AMA Ed Hub
UnderstandHealthcare-specific

Curated AI CME collection. Currency varies by item.

Cost
Not listed
Duration
Not listed
Level
Not listed
Cert / CME
CME credits (per page)
More details
Format
Self-paced
Code
N
HC
Y
Depth
U

Resource-list status: V
Listed under: Track 0–1: Orientation and AI-literate clinician (prompting, tools, PHI, no-code)

Link not machine-checked · Details researched: 26 Sep 2026Report a problem
#7
Medical Educator's Guide to Projects Leveraging AI and Learning Analytics
AMA ChangeMedEd
LeadHealthcare-specific

How to scope and partner on AI and learning-analytics projects in medical education.

Cost
Free
Duration
Not listed
Level
Not listed
Cert / CME
Not listed
More details
Format
PDF
Code
N
HC
Y
Depth
L (educators)

Resource-list status: V
Listed under: Track 0–1: Orientation and AI-literate clinician (prompting, tools, PHI, no-code)

Link checked: 2026-09-27 · Details researched: 26 Sep 2026Report a problem
#8
Making AI Work for You
Institute for Healthcare Improvement (IHI)
CourseUseLeadLLM-eraHealthcare-specific

Gen AI for quality-improvement use cases (run charts, Pareto, change ideas); risks and policy. Requires access to Copilot/ChatGPT etc. LLM+.

Cost
Not listed
Duration
8 weeks
Level
Needs some gen-AI experience
Cert / CME
Not listed
More details
Format
Online course
Code
N
HC
Y
Depth
Us, L

Resource-list status: V
Listed under: Track 0–1: Orientation and AI-literate clinician (prompting, tools, PHI, no-code)

Link checked: 2026-09-27 · Details researched: 26 Sep 2026Report a problem
#9
Artificial Intelligence (HMS Pre-Med online course)
Harvard Medical School Pre-Med
UnderstandHealthcare-specific

How AI detects patterns; imaging, signals, labs and EHR; therapy selection; AI in drug discovery; regulation and accountability; capstone clinical scenario. Faculty includes Arjun (Raj) Manrai.

Cost
Not listed
Duration
Not listed
Level
Beginner
Cert / CME
Not listed
More details
Format
Online
Code
N
HC
Y
Depth
U

Resource-list status: V
Listed under: Track 0–1: Orientation and AI-literate clinician (prompting, tools, PHI, no-code)

Link checked: 2026-09-27 · Details researched: 26 Sep 2026Report a problem
#10Link not machine-checked
AI knowledge repository (incl. NHS AI Lab case studies 2020–2025)
NHS England Digital
UnderstandLeadHealthcare-specific

Adoption best practice, spread/scale guidance, A–Z resources, AI Ethics Initiative. Warns content "may not be fully up to date".

Cost
Free
Duration
Not listed
Level
All
Cert / CME
Not listed
More details
Format
Web resource
Code
N
HC
Y
Depth
U, L

Resource-list status: V
Listed under: Track 0–1: Orientation and AI-literate clinician (prompting, tools, PHI, no-code)

Link not machine-checked · Details researched: 26 Sep 2026Report a problem
#11
AI in Clinical Decision Support & Diagnostics
Not listed
UnderstandEvaluateHealthcare-specific

CDS, imaging AI, predictive analytics, bias and ethics. Dated Dec 2025.

Cost
Coursera subscription
Duration
6 modules
Level
Not listed
Cert / CME
Certificate (Coursera)
More details
Format
Self-paced
Code
N
HC
Y
Depth
U, E

Resource-list status: V (provider field blank)
Listed under: Track 0–1: Orientation and AI-literate clinician (prompting, tools, PHI, no-code)

Link checked: 2026-09-27 · Details researched: 26 Sep 2026Report a problem
#85
Artificial Intelligence in Healthcare
E&ICT Academy, IIT Kanpur (EICTA consortium, MeitY)
UseEvaluateIndia

Clinical datasets, model evaluation, SHAP/LIME explainability, ChatGPT use cases, FHIR/HL7, ethics; capstone pitch. LLM-aware.

Cost
Not listed
Duration
35+ h
Cert
EICTA certificate (not IITK alumni status)
More details
Format
Live online
Code
N
Track / Depth
T1/T3, Us–E

Resource-list status: V
Listed under: India-specific courses (cross-track)

Link checked: 2026-09-27 · Details researched: 26 Sep 2026Report a problem
#91
Certificate Course on Applications of Artificial Intelligence in Health Sciences
University of Delhi
India

Eight modules: Introduction to AI in Healthcare; Introduction to Bioinformatics and Multi-Omics; Machine Learning Foundations in Medical Sciences; Medical Imaging and Data Handling; Deep Learning and CNNs for Medical Image Analysis; AI in Drug Discovery and Cheminformatics; Natural Language Processing in Healthcare; Ethics, Regulation, and Future of AI in Medicine. Eligible degrees include Medicine (MBBS), Nursing and Pharmacy.

Cost
Not listed
Duration
15 weeks; 60 learning hours (first batch: commencement 28 March 2026, tentative completion 19 July 2026)
Cert
Not listed
More details
Format
Online; batch size 50; level: Basic
Code
Not listed
Track / Depth
T1

Resource-list status: V (University of Delhi brochure, checked 2026-09-27; row rewritten on owner instruction)
Listed under: India-specific courses (cross-track)

Link checked: 2026-09-27 · Details researched: 26 Sep 2026Report a problem
#94
Deep Dive into LLMs like ChatGPT (single video)
Andrej Karpathy
VideoUnderstandLLM-eraGeneral AIFree (addendum)

Full training stack for a general audience: pretraining data, tokenisation, post-training, hallucinations and their mitigations, RL and "thinking" models.

Type
video
Year
2025-02
Duration/length
3 h 31 min
Level
Beginner
Currency
LLM-era
More details
Coding
N
HC-specific
General
Track/node
T0 / GenAI; T1 / hallucination
Depth
Understand

Resource-list status: S (length confirmed by Karpathy's own X post; ID seen in a secondary listing)
Listed under: Track 0: Free resources (addendum, verified 26 Sep 2026)

Link checked: 2026-09-27 · Details researched: 2026-09-26Report a problem
#98
Prompt engineering overview
Anthropic (Claude Platform Docs)
Official docUseLLM-eraGeneral AIFree (addendum)

Explains when prompting is the right tool, and says to define success criteria and tests before iterating. Links to guardrail pages ("Reduce hallucinations", "Increase output consistency").

Type
official doc
Year
2026 (continuously updated)
Duration/length
multi-page
Level
Beginner
Currency
LLM-era
More details
Coding
N
HC-specific
General
Track/node
T1 / prompting
Depth
Use

Resource-list status: V
Listed under: Track 1: Free resources (addendum, verified 26 Sep 2026)

Link checked: 2026-09-27 · Details researched: 2026-09-26Report a problem
#99
Prompting best practices
Anthropic
Official docUseLLM-eraGeneral AIFree (addendum)

Clarity, examples, XML structure, "quote-then-answer" grounding, and thinking and agentic prompting for current models.

Type
official doc
Year
2026
Duration/length
long page
Level
Beginner–Int
Currency
LLM-era
More details
Coding
N
HC-specific
General (includes a worked clinical example: quote extraction to support a doctor's diagnosis)
Track/node
T1 / prompting; hallucination & verification
Depth
Use

Resource-list status: V
Listed under: Track 1: Free resources (addendum, verified 26 Sep 2026)

Link checked: 2026-09-27 · Details researched: 2026-09-26Report a problem
#100
Interactive Prompt Engineering Tutorial (9 chapters + appendix)
Anthropic
GitHubNotebookUseLLM-eraGeneral AIFree (addendum)

Hands-on exercises with answer keys covering prompt structure, roles, examples, chaining and avoiding hallucination.

⚠ Check currency — the resource list flags this entry (see the fields marked ⚠).

Type
GitHub / notebook
Year
2024
Duration/length
9 chapters
Level
Beginner
Currency
LLM-era, but written for Claude 3 Haiku ⚠ (techniques still apply)
More details
Coding
Opt (Jupyter or Google Sheets)
HC-specific
General
Track/node
T1 / prompting
Depth
Use

Resource-list status: V
Listed under: Track 1: Free resources (addendum, verified 26 Sep 2026)

Link checked: 2026-09-27 · Details researched: 2026-09-26Report a problem
#101
Anthropic Academy (free courses incl. AI Fluency, Claude 101)
Anthropic
CourseUseLLM-eraGeneral AIFree (addendum)

Official free courses with certificates. The AI Fluency tracks suit clinicians and educators.

⚠ Check currency — the resource list flags this entry (see the fields marked ⚠).

Type
course hub
Year
2025–2026
Duration/length
self-paced, 1–3 h per course
Level
Beginner
Currency
LLM-era
More details
Coding
N
HC-specific
General
Track/node
T1 / AI tool literacy
Depth
Use

Resource-list status: V ⚠ free email registration required
Listed under: Track 1: Free resources (addendum, verified 26 Sep 2026)

Link checked: 2026-09-27 · Details researched: 2026-09-26Report a problem
#102
Anthropic Learn hub
Anthropic
Official docUseLLM-eraGeneral AIFree (addendum)

Free courses, tutorials and use cases from Anthropic's education team, including AI Fluency courses and an Introduction to Model Context Protocol (MCP) course.

Now called Claude Academy (the Anthropic Learn address redirects here). Its MCP courses are why it is also listed on Track 6.

Type
official doc
Year
2026
Duration/length
hub
Level
Beginner
Currency
LLM-era

Official page (found 2026-09-28): https://academy.claude.com/

More details
Coding
N
HC-specific
General
Track/node
T1 / tool literacy
Depth
Use

Resource-list status: V
Listed under: Track 1: Free resources (addendum, verified 26 Sep 2026)

Link checked: 2026-09-28 · Details researched: 2026-09-26Report a problem
#138· also listed as #83
AI for clinicians (noc26_ge96)
IIT Madras — Balaji Srinivasan, Ganapathy Krishnamurthi
MOOCUnderstandUseLLM-eraHealthcare-specificIndiaFree (addendum)

The only NPTEL course found that is explicitly about AI in healthcare. Taught by an engineering and medical-imaging team.

Type
MOOC (free lectures)
Year
2026
Duration/length
weeks not confirmed
Level
Beginner
Currency
LLM-era run
More details
Coding
Opt
HC-specific
HC
Track/node
T0/T1 India anchor
Depth
Understand/Use

Resource-list status: S
Listed under: India cross-cutting — NPTEL/SWAYAM (all free to audit; certification optional) — free addendum, verified 26 Sep 2026

Link checked: 2026-09-27 · Details researched: 2026-09-26Report a problem
#170
Large language models encode clinical knowledge
Singhal K et al.
PaperUnderstandLLM-eraOpen access

Med-PaLM and MultiMedQA; first peer-reviewed passing-level medical QA plus a clinician-rated harm framework

Journal
Nature
Year
2023
Paper type
Model development / validation

Free full text (Unpaywall, 2026-09-27): https://www.nature.com/articles/s41586-023-06291-2.pdf

Read this critically → the 15-question checklist

More details
Open-access status
Not confirmed
Track and node
T1 · What LLMs know
Depth level
Understand
Currency note
LLM-era; model superseded, framework still used

Resource-list status: V
Listed under: Track 1: Papers — AI-literate clinician (LLMs in practice)

Link checked: 2026-09-27 · Details researched: 2026-09-27Report a problem
#171
Large language models in medicine
Thirunavukarasu AJ et al.
PaperUnderstandLLM-eraPaywalled

Clinician-oriented explanation of how LLMs are built, where they help and fail

Journal
Nature Medicine
Year
2023
Paper type
Narrative review

Read this critically → the 15-question checklist

More details
Open-access status
Not confirmed
Track and node
T1 · LLM basics
Depth level
Understand
Currency note
LLM-era; authoritative

Resource-list status: V
Listed under: Track 1: Papers — AI-literate clinician (LLMs in practice)

Link checked: 2026-09-27 · Details researched: 2026-09-27Report a problem
#172
Benefits, Limits, and Risks of GPT-4 as an AI Chatbot for Medicine
Lee P et al.
PaperUnderstandLLM-eraPaywalled

Worked examples of GPT-4 errors; the reference text for "always verify"

Journal
NEJM
Year
2023
Paper type
Special report

Read this critically → the 15-question checklist

More details
Open-access status
Not confirmed
Track and node
T1 · Hallucination & verification
Depth level
Understand
Currency note
LLM-era; lessons hold

Resource-list status: V
Listed under: Track 1: Papers — AI-literate clinician (LLMs in practice)

Link checked: 2026-09-27 · Details researched: 2026-09-27Report a problem
#173
Comparing Physician and Artificial Intelligence Chatbot Responses to Patient Questions Posted to a Public Social Media Forum
Ayers JW et al.
PaperEvaluateLLM-eraOpen access

Chatbot answers rated higher for quality and empathy than physicians'; triggered the patient-messaging debate

Journal
JAMA Internal Medicine
Year
2023
Paper type
Cross-sectional

Free full text (Unpaywall, 2026-09-27): https://www.ncbi.nlm.nih.gov/pmc/articles/10148230

Read this critically → the 15-question checklist

More details
Open-access status
Not confirmed
Track and node
T1 · Patient-facing use
Depth level
Evaluate
Currency note
LLM-era; design limits (Reddit sample)

Resource-list status: V
Listed under: Track 1: Papers — AI-literate clinician (LLMs in practice)

Link checked: 2026-09-27 · Details researched: 2026-09-27Report a problem
#174
Accuracy and Reliability of Chatbot Responses to Physician Questions
Goodman RS et al.
PaperEvaluateLLM-eraOpen access

Multispecialty physician questions: mostly accurate, but with confident errors

Journal
JAMA Network Open
Year
2023
Paper type
Cross-sectional

Free full text (Unpaywall, 2026-09-27): https://jamanetwork.com/journals/jamanetworkopen/articlepdf/2809975/goodman_2023_oi_231053_1695240536.8461.pdf

Read this critically → the 15-question checklist

More details
Open-access status
Likely (OA journal)
Track and node
T1 · Hallucination & verification
Depth level
Evaluate
Currency note
LLM-era

Resource-list status: V
Listed under: Track 1: Papers — AI-literate clinician (LLMs in practice)

Link checked: 2026-09-27 · Details researched: 2026-09-27Report a problem
#175
Capabilities of GPT-4 on Medical Challenge Problems
Nori H et al.
PaperUnderstandLLM-eraOpen access

General model exceeded USMLE pass mark without medical fine-tuning

Journal
arXiv preprint
Year
2023
DOI
Not confirmed
Paper type
Benchmark

DOI (confirmed 2026-09-27): https://doi.org/10.48550/arXiv.2303.13375

Free full text (Unpaywall, 2026-09-27): https://arxiv.org/pdf/2303.13375

Read this critically → the 15-question checklist

More details
Open-access status
Likely (preprint)
Track and node
T1 · Exams & benchmarks
Depth level
Understand
Currency note
LLM-era; preprint only

Resource-list status: S
Listed under: Track 1: Papers — AI-literate clinician (LLMs in practice)

Link checked: 2026-09-27 · Details researched: 2026-09-27Report a problem
#176
Performance of ChatGPT on USMLE: potential for AI-assisted medical education using large language models
Kung TH et al.
PaperUnderstandLLM-eraOpen access

First widely publicised ChatGPT-passes-USMLE study

Journal
PLOS Digital Health
Year
2023
DOI
Not confirmed
Paper type
Benchmark

DOI (confirmed 2026-09-27): https://doi.org/10.1371/journal.pdig.0000198

Free full text (Unpaywall, 2026-09-27): https://journals.plos.org/digitalhealth/article/file?id=10.1371/journal.pdig.0000198&type=printable

Read this critically → the 15-question checklist

More details
Open-access status
Likely (OA journal)
Track and node
T1 · Exams & benchmarks
Depth level
Understand
Currency note
LLM-era; historical marker

Resource-list status: S
Listed under: Track 1: Papers — AI-literate clinician (LLMs in practice)

Link checked: 2026-09-27 · Details researched: 2026-09-27Report a problem
#177
Accuracy of a Generative Artificial Intelligence Model in a Complex Diagnostic Challenge
Kanjee Z et al.
PaperEvaluateLLM-eraOpen access

GPT-4 on NEJM CPC cases; baseline for LLM differential diagnosis

Journal
JAMA
Year
2023
DOI
Not confirmed
Paper type
Research letter

DOI (confirmed 2026-09-27): https://doi.org/10.1001/jama.2023.8288

Free full text (Unpaywall, 2026-09-27): https://www.ncbi.nlm.nih.gov/pmc/articles/10273128

Read this critically → the 15-question checklist

More details
Open-access status
Not confirmed
Track and node
T1 · Diagnostic reasoning
Depth level
Evaluate
Currency note
LLM-era

Resource-list status: S
Listed under: Track 1: Papers — AI-literate clinician (LLMs in practice)

Link checked: 2026-09-27 · Details researched: 2026-09-27Report a problem
#178
Large Language Model Influence on Diagnostic Reasoning: A Randomized Clinical Trial
Goh E et al.
PaperEvaluateLLM-eraOpen access

LLM access did not improve physicians' diagnostic reasoning, though the LLM alone scored well

Journal
JAMA Network Open
Year
2024
DOI
Not confirmed
Paper type
RCT

DOI (confirmed 2026-09-27): https://doi.org/10.1001/jamanetworkopen.2024.40969

Free full text (Unpaywall, 2026-09-27): https://pmc.ncbi.nlm.nih.gov/articles/instance/11519755/bin/jamanetwopen-e2440969-s001.pdf

Read this critically → the 15-question checklist

More details
Open-access status
Likely (OA journal)
Track and node
T1 · Human–AI teaming
Depth level
Evaluate
Currency note
LLM-era; important negative RCT

Resource-list status: S
Listed under: Track 1: Papers — AI-literate clinician (LLMs in practice)

Link checked: 2026-09-27 · Details researched: 2026-09-27Report a problem
#179
GPT-4 assistance for improvement of physician performance on patient care tasks: a randomized controlled trial
Goh E et al.
PaperEvaluateLLM-eraOpen access

LLM assistance improved management reasoning; counterpoint to the 2024 diagnostic RCT

Journal
Nature Medicine
Year
2025
Paper type
RCT

Free full text (Unpaywall, 2026-09-27): https://europepmc.org/articles/PMC12380382

Read this critically → the 15-question checklist

More details
Open-access status
Not confirmed
Track and node
T1 · Human–AI teaming
Depth level
Evaluate
Currency note
LLM-era; publisher correction reported (10.1038/s41591-025-03586-x, seen secondarily)

Resource-list status: V
Listed under: Track 1: Papers — AI-literate clinician (LLMs in practice)

Link checked: 2026-09-27 · Details researched: 2026-09-27Report a problem