Chatbots beat doctors on paper. Then real patients used them and it fell apart.

Run time: 6:50

In today's episode:

1. Chatbot answers rated more empathetic than doctors, preferred 79% of the time
2. GPT-4 alone beat physicians by ~15 points on diagnostic vignettes
3. Google's AMIE matched or beat primary care docs in text consultations
4. Microsoft's system solved 85% of hard NEJM cases vs doctors' 20%
5. Oxford RCT: patients using LLMs did no better than a web search
6. Half of consumer chatbot medical advice carried potential for harm
7. ~1 in 6 US adults now use an AI chatbot for health advice monthly

TL;DR:

• On clean written cases, AI genuinely out-diagnoses doctors — that part is peer-reviewed, not hype.
• On real patients doing their own typing, that edge vanishes: an Oxford RCT (~1,300 people) found LLM users did no better than web search, because patients omit the detail that matters and models mix right advice with wrong.
• What's actually deployed is the opposite of a doctor replacement: clinician tools like OpenEvidence (used by ~65% of US doctors) keep a licensed human between the model and the prescription. No LLM is FDA-cleared to diagnose you solo.

Sources cited:

• JAMA Internal Medicine, Ayers et al.
• JAMA Network Open, Goh et al.
• Nature, Google DeepMind
• Microsoft AI
• Oxford, Nuffield Dept of Primary Care (Payne et al.)
• NBC News on OpenEvidence

Subscribe: YouTube

medAI Times is for educational and informational purposes only. The content does not constitute medical advice, diagnosis, treatment recommendation, or professional clinical guidance. Consult qualified healthcare professionals and refer to official sources before making clinical, research, regulatory, or business decisions.

Three hundred million health questions a week. Chat G P T just opened the doors.

Run time: 7:11

In today's episode:

1. ChatGPT Health opens to every US adult
2. FDA lets Dexcom skip part of premarket review
3. White House starts clinical AI benchmarking sprint
4. Nature Medicine RCT: eye AI lifts diagnosis
5. Utah publishes AI prescription renewal data
6. Medtronic clears real-time AI for surgery
7. Tempus buys Personalis for $1.5 billion
8. Epic's AI leader joins an executive exodus
9. AMD invests up to $5 billion in Anthropic
10. Anthropic opens rare-disease research grants
11. OpenAI's test models breached Hugging Face
12. Google ships Gemini 3.6 Flash, still no 3.5 Pro

TL;DR:

• ChatGPT Health is now open to every US adult on every plan, connected to Epic and Oracle Health records, at 300 million health queries a week — with zero outcome evidence and a fresh lawsuit.
• Washington moved twice in one week: the FDA gave Dexcom the first premarket exemption under its TEMPO pilot in exchange for Medicare-generated real-world evidence, while OSTP, FDA and ONC opened a one-month sprint to define how clinical AI should be evaluated at all.
• The week's only top-tier randomized trial (Nature Medicine, July 24) worked because a clinician stayed in the loop: AI-assisted top-5 genetic accuracy 88.5% vs 67.3% for controls in inherited retinal disease.

Sources cited:

• TechCrunch
• MedTech Dive
• STAT
• Nature Medicine
• Forbes
• Medtronic
• MedCity News
• STAT
• AMD/GlobeNewswire
• Anthropic

Subscribe: YouTube

medAI Times is for educational and informational purposes only. The content does not constitute medical advice, diagnosis, treatment recommendation, or professional clinical guidance. Consult qualified healthcare professionals and refer to official sources before making clinical, research, regulatory, or business decisions.

Two hundred million protein shapes, one Nobel Prize, and still zero A I drugs approved.

Run time: 6:32

In today's episode:

1. AlphaFold two hit ~90/100 accuracy at CASP14 (2020)
2. Database now covers 214M+ protein sequences, 3M+ users
3. 2024 Nobel Prize: Hassabis, Jumper (AlphaFold) + Baker (design)
4. AlphaFold3 predicts drug-protein binding, 58% vs 24% docking
5. Static shapes only — fails on flexible, shape-shifting targets
6. Zero AlphaFold-designed drugs approved; first trials targeted end-2026

TL;DR:

• Real, at the research layer: AlphaFold solved a 50-year protein-shape problem (peer-reviewed, Nobel-confirmed) and put 214M+ free structures in front of 3M+ scientists in ~190 countries — it changed how medical research begins.
• Not yet, at the clinic: No AlphaFold-designed drug is approved or, as of spring 2026, even dosed in a patient; the models predict one static shape and stumble on the flexible drug targets that matter most.
• The milestone to watch: Isomorphic Labs (DeepMind spinout, $2.1B Series B) targets its first human cancer trials by end-2026 — that first-dose event, not the Nobel, is where "changed medicine" gets tested against outcome evidence.

Sources cited:

• Nature 2021, Jumper et al.
• Nucleic Acids Research 2024, Varadi et al.
• Nature 2024, Abramson et al.
• EMBL-EBI resource update
• PMC 2025, anti-schistosomal docking pipeline
• bioRxiv 2025, AlphaFold3 in Drug Discovery: capabilities & limitations
• Isomorphic Labs, $2.1B Series B (May 2026)
• NobelPrize.org, Chemistry 2024

Subscribe: YouTube

medAI Times is for educational and informational purposes only. The content does not constitute medical advice, diagnosis, treatment recommendation, or professional clinical guidance. Consult qualified healthcare professionals and refer to official sources before making clinical, research, regulatory, or business decisions.

Mayo Clinic is now running a hundred and fifty AI models at once.

Run time: 6:48

In today's episode:

1. Mayo Clinic now runs 150 AI models
2. Senate probes Medicare Advantage AI care denials
3. Specialized tool beats GPT, Gemini, Claude
4. Causal AI: the vasopressor call decides survival
5. AI drugs pass safety, stall on efficacy
6. NeuroVFM reads a hospital's own scans
7. Claude adds self-serve HIPAA setup
8. Moonshot's Kimi K3 lands at 2.8T params

TL;DR:

• Mayo's 150 models plus a whistleblower suit show hospital AI's real fight has moved from capability to governance.
• The Senate and the HHS inspector general are closing in on Medicare Advantage AI denials, and the GOP just blocked a bid to kill the WISeR AI prior-auth pilot.
• Three peer-reviewed reads this week — OpenEvidence beating frontier LLMs, causal sepsis dosing, and the AI-drug Phase 2 wall — all land on the same point: a benchmark is not a bedside outcome.

Sources cited:

• CNN Business (via KESQ)
• STAT
• Healthcare Dive
• STAT (WISeR vote)
• arXiv 2606.28960
• medRxiv 10.64898/2026.07.06.26357375
• Drug Discovery Today (2024)
• Nature Medicine
• Anthropic (release notes via Releasebot)
• LLM-Stats AI news

Subscribe: YouTube

medAI Times is for educational and informational purposes only. The content does not constitute medical advice, diagnosis, treatment recommendation, or professional clinical guidance. Consult qualified healthcare professionals and refer to official sources before making clinical, research, regulatory, or business decisions.

Three countries claim they're winning healthcare AI. None can prove it helps patients yet.

Run time: 6:44

In today's episode:

1. US leads on approvals: 1,400-plus FDA-cleared AI devices
2. China leads on scale: DeepSeek live in 260-plus hospitals
3. UAE leads on data: ~1M genomes sequenced, open clinical LLM
4. Deep Knowledge index (May 2026): US 1, UK 2, China 3
5. The catch: almost none of it tied to patient outcomes

TL;DR:

• There's no single winner — the US leads on capital, approvals, and frontier models; China leads on national-scale deployment; the UAE leads on genomic depth and sovereign data. Different races, different scoreboards.
• The headline numbers are inputs, not outcomes: FDA device counts, hospital deployments, genomes sequenced. None of the three has published randomized, outcome-level evidence that its AI makes patients live longer.
• China's flagship "Agent Hospital" 93% score is a simulation (AI treating AI-generated patients); America's ~1,400 clearances mostly ran the 510(k) resemblance pathway; the UAE's genome bank has few published clinical endpoints so far.

Sources cited:

• FDA AI-Enabled Medical Devices
• CKGSB Knowledge
• Global Times
• Semafor
• medRxiv preprint
• G42 / M42
• SCMP / Deep Knowledge Group

Subscribe: YouTube

medAI Times is for educational and informational purposes only. The content does not constitute medical advice, diagnosis, treatment recommendation, or professional clinical guidance. Consult qualified healthcare professionals and refer to official sources before making clinical, research, regulatory, or business decisions.

OpenAI just signed eight hospital systems the same week two studies said not so fast.

Run time: 6:29

In today's episode:

1. OpenAI signs eight major hospital systems
2. NHS deploys AI scribes to every A&E
3. Nature Medicine finds top medical models brittle
4. Kenya trial: AI helped notes, not patients
5. Weill Cornell agents design clinical trials
6. Cardiac digital twin flags non-responders before implant
7. Philips gets FDA nod for AI ultrasound
8. Medicare launches new health-tech and AI office
9. Claude Cowork expands to web and mobile

TL;DR:

• OpenAI is buying its way into hospitals — eight health systems signed, 230M weekly health users — but it's distribution, not clinical evidence.
• Two peer-reviewed papers this week undercut the hype: Nature Medicine shows flagship models are brittle under simple adversarial tweaks, and a Kenyan RCT shows an AI helper improved notes but not patient outcomes.
• The at-scale wins remain operational, not diagnostic: NHS ambient scribes save ~47 min/shift; Philips' new FDA-cleared ultrasound automates measurement, not diagnosis.

Sources cited:

• Forbes
• NHS England
• Nature Medicine
• Nature Medicine
• Weill Cornell
• medRxiv
• Philips
• Healthcare Dive
• TechCrunch
• Axios

Subscribe: YouTube

medAI Times is for educational and informational purposes only. The content does not constitute medical advice, diagnosis, treatment recommendation, or professional clinical guidance. Consult qualified healthcare professionals and refer to official sources before making clinical, research, regulatory, or business decisions.

Ten years ago an AI pioneer said stop training radiologists. The field grew instead.

Run time: 5:40

In today's episode:

1. Hinton's 2016 "stop training radiologists" call has now expired
2. On narrow tasks, AI beats radiologists on average (AUC 0.88)
3. A workforce model projects a ~33% cut in radiologist hours over 5 years
4. But models crater on external data and disagree with each other
5. Human + AI beats AI alone in every head-to-head
6. Autonomy is live only for auto-clearing normal chest X-rays
7. Verdict: the job is being rearranged, not deleted

TL;DR:

• The strong claim ("AI replaces radiologists") is pundit/vendor tier; the peer-reviewed and prospective data consistently show human-plus-AI beating AI alone, with autonomy confined to triaging normal studies.
• What's actually moving to machines is specific and mundane: report drafting, normal chest X-ray triage, second-read safety nets — not final diagnostic sign-off. Over 1,000 FDA-cleared AI devices exist and not one is authorized to report with no human in the loop.
• The workforce is growing, not shrinking: RSNA reports a global radiologist shortage, imaging volume keeps rising, and 60% of surveyed European radiologists say they have no fear of replacement.

Sources cited:

• Multi-target AI, 16 findings on chest/abdominal CT
• The Effect of AI on the Radiologist Workforce: A Task-Based Analysis
• Systematic review of AI generalizability across clinical settings
• Seven commercial lung-cancer CXR tools head-to-head, Radiology (RSNA)
• Standalone AI vs AI-assisted radiologists for intracranial hemorrhage
• Nature Medicine, "How to meaningfully evaluate AI in clinical medicine"
• India multicenter QI
• UK service evaluation
• The role of AI in mitigating radiologist shortages (systematised review)
• NEJM AI automation-bias RCT

Subscribe: YouTube

medAI Times is for educational and informational purposes only. The content does not constitute medical advice, diagnosis, treatment recommendation, or professional clinical guidance. Consult qualified healthcare professionals and refer to official sources before making clinical, research, regulatory, or business decisions.

The most powerful AI the US banned last month is back — and biomedical labs want it.

Run time: 5:33

In today's episode:

1. US lifts export ban; Fable 5, Mythos 5 back worldwide
2. Anthropic launches Claude Science for pharma research
3. Claude Sonnet 5 ships: near-Opus quality, cheaper
4. FDA clears on-premise brain-tumor MRI AI (Cercare)
5. GE HealthCare clears auto-contouring for radiation therapy
6. Feds reprimand AI Medicare prior-auth pilot (WISeR)
7. Qualified Health raises $125M for hospital AI
8. NVIDIA's surgical robotics foundation models advance

TL;DR:

• The US reversed its June 12 export ban: Anthropic's Fable 5 and Mythos 5 are back worldwide as of July 1 with a new safety classifier — restoring the drug-design model biomedical labs relied on. Access restored, not efficacy proven.
• Anthropic doubled down on medicine the same week: Claude Science (a research workbench aimed at pharma) and Claude Sonnet 5 (near-Opus quality at $2/$10 per Mtok) both launched June 30.
• Two FDA imaging clearances (Cercare's on-premise brain-tumor segmentation; GE's radiation-therapy auto-contouring) and one federal reprimand of an AI Medicare prior-auth contractor round out the week — capability moving forward, oversight catching up.

Sources cited:

• Al Jazeera
• STAT
• AuntMinnie
• AuntMinnie
• KUOW
• Fierce Healthcare
• Anthropic
• Axios
• NVIDIA / Hugging Face

Subscribe: YouTube

medAI Times is for educational and informational purposes only. The content does not constitute medical advice, diagnosis, treatment recommendation, or professional clinical guidance. Consult qualified healthcare professionals and refer to official sources before making clinical, research, regulatory, or business decisions.