AI is eroding judgment on both sides of the screen

Two new studies, one on how AI models fail under a false sentence, one on how we fail under a confident answer. Read together, they're a warning.

Antidepressants: neither miracle nor menace

Ten common claims about antidepressants, and what the evidence actually says about each.

AI amplifies whatever the system already rewards

Two arguments, one on organizational readiness, one on primary care, that land on the same uncomfortable truth.

We're watching the wrong side of the AI-in-therapy debate

Two new studies, one on what clinicians actually do with AI, one on whether it works, redraw the whole conversation.

Reasoning beats recall

An 8-billion-parameter agent out-reasoned models 80x its size at choosing treatments, and the reason is the most useful lesson in medical AI right now.

AI is failing the way doctors fail

A Mayo Clinic team stopped scoring AI's answers and started scoring its reasoning. What they found should change how we test clinical AI.

75% believe in AI. 6% are ready for it.

Why buying tools isn't the same as being able to run them, and the five components that separate the two.

"Physicians Rated It Higher Than Physicians" - A Claim Worth Questioning

"OpenAI's health claims sound impressive. But are we measuring the right things?"

We train physicians to practice, not to lead

The leadership gap in medical training, what I learned inside hospital administration, and a model for closing it.

The Clinical error that AI keeps making (And We're Not Catching It)

LLMs frequently commit to answers when they should pause, ask, or escalate. It's premature closure, and it's costing clinical validation.

A small evolution (and what's coming next)

Same newsletter. Broader lens. Here's why.

Training didn't stop it: Why physicians still trust flawed AI

44 physicians with formal AI training still made worse decisions when exposed to erroneous AI recommendations.

$63 billion invested in AI health startups. Here's where it went.

Spoiler: Not where clinical need is highest.

AI in healthcare just cost companies $5.7 billion

Five cases. Same pattern. No audit trails.

ChatGPT treats psychotic crises like fun mysteries to solve

The addiction field has been waiting 30 years for this

Which psychiatric medications actually reduce suicide risk?

New evidence from 6M+ patients.

If AI can replace radiologists, why can't it replace hospital CEOs first?

NYC hospital CEO wants AI to replace radiologists.

AI scribes have a 70% error rate.

We're scaling them anyway.

LLMs recommended more opioids for marginalized patients while flagging them as higher addiction risk.

10 LLMs tested on 1,000 pain vignettes. The bias patterns are unacceptable.

89% of physicians 60+ are using AI or are interested. Age isn't a barrier.

Doximity surveyed 3,151 physicians. Adoption is widespread, not generational.

AI beats doctors in synthetic scenarios. Real clinical practice is messier.

Benchmark studies use complete data and clear criteria. Medicine doesn't work that way.

Stanford-Harvard predicts clinical AI's future.

Three predictions should worry you.

Using AI to validate AI?

You're measuring fiction, not performance.