Colin Son, MD

the brain is the only organ where we've collectively decided to wait for catastrophe before paying attention to it no covered cognitive baseline at 45. no reimbursed brain MRI for high-risk asymptomatics. no neuropsych eval before the first symptom we built entire preventive infrastructure around the heart and colon. the organ running everything else gets nothing until it fails

The weirdest thing about healthcare AI policy is it only imagines two futures: nothing changes, or robots replace all doctors. Nobody is planning for the messy middle where AI makes mediocre doctors good and good doctors fast.

@gregorykennedy The exchange sucks it’s like 10000 RMB or something

Somehow, I became a LinkedIn millionaire, but like, I still feel poor?

Cool, cool

@WillyRontgen Cool, cool

How it started vs how it’s going. See too many of these unfortunately. Would avoid spine surgery at all costs.

@sierracatalina How can you tell?

DOWN DETECTOR IS DOWN .

I mean I wish to give no more dollars than I need to to China But…. Minimax M2.7 is very good as an agent. And as importantly lets me use my $40 subscription in Hermes. Rather than $400 in anthropic API calls.

The most tragic irony of modern medicine is a physician spending 11 minutes on documentation for every 1 minute of patient contact, then complaining AI is "impersonal."

Welcome to America where we know how to optimize geriatric health!

@CrainsNewYork @RadiologyBiz @NYCHealthSystem When you want an AI to read a mammogram: → FDA SaMD gauntlet → 510(k) or De Novo → Clinical trials → Post-market surveillance → Massive liability When you want an AI to replace a hospital CEO? Nothing. Zero FDA. Zero regulators. Zero. Days. To. Deploy.

Every scientific paper in 2026 is still uploaded as a formatted PDF to a site that rate-limits downloads. We built AI that can fold proteins but we can't share a research paper as structured data.

GLP-1 drugs are speedrunning every indication in medicine. Weight loss, heart failure, addiction, psoriatic arthritis. At this point I'm waiting for the trial showing they fix hospital WiFi.

@thatguybg That rizz. Damn

anthropic's mythos benchmarks have been leaked its clear this is the first model that truly threatens human civilization.

This is so tech bubble Like 10-25% of white collar workers themselves view their jobs as meaningless and it’s even higher than that in terms of who really contributes I agree you explain why high earners work more, but it doesn't refute that a meaningful share of total employment is pointless, or that efficiency gains per person has slowed

“fake work” and “bullshit jobs” has been fantastically wrong and misleading for understanding the modern world. a much better understanding is of a global economy where minor skill differences and improvements lead to monumentally different outcomes, and the marginal hour of work has never been more measurable or useful after the advent of even moderately effective talent allocation systems and the variability of reward based on effort and skill, people have engaged much harder in a red queen rat race across the world. this is why the Chinese ‘cram schools’ exist and why ‘yuppie striverism’ is a thing and why people trade off later family formation for working more so often. while overall work hours are slightly down, they are actually up for high earners (nber.org/digest/jul06/w…) I see it in the marginal effect with my friends now after the advent of claude and codex: they are actually working harder now than they ever have before. this is due to a personal Jevon’s paradox where they see that the value of their time has increased dramatically, that they can get a lot more visible work done towards goals they care about than they used to after requests from their customers the labs are doing things like inventing dispatch which lets you monitor work and manipulate your computer from your phone, on top of prior changes like having always on communications (slack). You hear about people launching codex jobs from their phone the moment they have an idea and reviewing them later no clue how long this lasts but the most immediate impact of co-existing with the machine state is higher productivity and higher visibility which leads to more work hours

CEO of America’s largest public hospital system says he’s ready to replace radiologists with AI. Mitchell H. Katz, MD, the president and CEO of New York City’s public hospital system, has stated that he is prepared to begin replacing radiologists with artificial intelligence for certain diagnostic tasks as soon as the regulatory environment permits. Speaking at a recent panel, Katz highlighted the potential for AI to serve as the "first reader" for routine screenings such as mammograms and X-rays, which would allow hospitals to realize significant financial savings amid the rising costs of human specialists. In this scenario, radiologists would transition into a secondary role, only reviewing images that the technology identifies as abnormal. Other healthcare leaders supported this vision, citing data that suggests AI can be more accurate than humans in specific low-risk screenings. However, this proposal has met with intense opposition from medical professionals who argue that such a shift would pose a severe threat to patient safety. Radiologists have criticized hospital administrators for being "confidently uninformed," claiming that current AI technology is not yet capable of providing independent patient care without human oversight.

Your heart rhythm determines how well your brain takes out its trash. New ACC data showing AFib reduces brain glymphatic flow. The heart-brain axis is going to rewrite how we think about neurodegeneration.

@mercebent Get out. 3 body got shit right

"which sci-fi books got the most predictions about technology and science right?"

@maxmarchione @cwhogg @FarvingCo I mean for something like pain there can be no other endpoint. And who gives a shit about the mechanism - placebo or physiologic - still relief. Just a matter of safety profile.

@cwhogg @FarvingCo Yeah placebo is insanely powerful, agreed. Though I wouldn’t go as far as to say that “patient reporting means nothing”. It’s an endpoint in many trials (placebo-controlled ofc)

11 of 12 patients reported SIGNIFICANT KNEE PAIN RELIEF after a SINGLE injection of BPC-157. Relief lasted 6-12 months. Chronic knee pain — osteoarthritis, meniscus tears, ligament sprains — who got one intra-articular injection. How long have you been managing your knee pain? How many cortisone shots have you gotten that wore off in weeks? How much ibuprofen are you taking just to train? 11 of 12 got one injection. Relief lasted up to a year. This is one of only three published human studies on BPC-157 in existence. 12 women with interstitial cystitis. Every one of them had failed standard treatment for over a year. 10 of 12 reported complete symptom resolution at six weeks. Doctors confirmed it with cystoscopic imaging — visible tissue healing. 2 healthy adults received IV BPC-157 at doses up to 20mg. Cardiac, liver, kidney, thyroid, and metabolic biomarkers: zero measurable effects. No adverse events. Plasma cleared within 24 hours. PMID 40131143 Three human studies. Zero adverse effects across all three. And a 2025 systematic review of 544 articles found consistently positive outcomes in every preclinical model — muscle, tendon, ligament, and bone. PMID 40756949 A pentadecapeptide isolated from human gastric juice. Native to your body. No lethal dose achieved in any study. Stable in stomach acid for 24+ hours. The science was never the problem. If you’re going to try oral BPC-157 — most of what’s sold online is underdosed garbage that degrades in stomach acid. Make sure it’s third-party tested and actually gastric stable. Most isn’t. PMID 34324435

The failure mode of AI in medicine isn't the AI being wrong. It's the AI confirming your wrong diagnosis with the confidence of a tenured attending.

If you don't have LLM psychosis you're not using LLMs ambitiously enough

@agazdecki Fuck

POV: You didn't become a millionaire after 1 week of leaving your full time job and building a startup.