Ming Jack Po MD PhD

@joshk deli inflation is brutal because they don't re-price all that often, so prices suddenly jump by 50% to 100%!

Inflation hits the West Village

I’ve been speaking with reporters all day about the firing of USPSTF leadership, and the stakes couldn’t be higher. The Task Force’s credibility depends on rigorous, independent review of evidence. We should all be paying attention to what happens next. axios.com/2026/05/20/ken…

I can't believe I'm out of tokens again across multiple AI vendors.. argh.. can someone make a 500 dollar plan to save me from myself?

I really did not want to be writing this morning. But this issue is too important. It’s been said that the last thing the weather person should do before going on the air is look out the window. The _day before_ @eringriffith profiled Medvi in @nytimes as the AI company of the future, the FDA approved Lilly’s new oral medication: orforglipron. *Not tirzepatide.* In fact, it’s not even a peptide. It’s the first non-peptide, small-molecule oral GLP-1 receptor agonist. Lilly owns tirzepatide. They invented it. If you could put tirzepatide in a pill, Lilly would do it. They would desperately want to. Instead, they spent millions and nearly eight years licensing a completely different kind of molecule, because oral tirzepatide is a biological impossibility. Tirzepatide is a 39-amino-acid, 4,813-dalton peptide. Your gut does not distinguish it from a piece of chicken. SNAC, the absorption enhancer that barely gets oral semaglutide to 1% bioavailability, is compound-specific. It failed with liraglutide, another GLP-1 peptide, and has no demonstrated mechanism for tirzepatide. There is no published human study of oral tirzepatide. There is no plausible mechanism. Medvi sells it starting at $279 a month. A RICO class action against its supply chain partners has already called the product modern-day snake oil. Lilly’s own strategy is the best witness. We can hear the lawyers now: “So doctor, what was your assumption on why Lilly was not pursuing oral tirzepatide despite that in not doing so they would instead pursue an entire entirely different type of molecule and possibly create market confusion with their new entrant?” The Times profile actually described an accountability architecture whose impact in part is that no single entity owns the patient and process. Medvi handles marketing. CareValidate provides the clinical workflow. OpenLoop provides prescribers and pharmacy fulfillment. The marketing layer can say the doctors make the decisions. The doctor platform can say the brand controls the messaging. The prescribers say the pharmacy fills what’s ordered. Everyone can point at everyone else. That structure explains a lot of the financials. Medvi reported a 16.2% net margin. Hims, with 2,442 employees selling the same drug categories, reported 5.5%. The 10.7-point spread represents in part everything Medvi may not pay for: extensive clinical oversight, advanced adverse event monitoring, satisfactory regulatory compliance, sound quality systems. The Times says they verified Medvi’s revenue. They did not seem to verify or note many other aspects. Six weeks before the profile ran, the FDA had issued Warning Letter #721455 for misbranding compounded GLP-1s. OpenLoop had disclosed a data breach: a threat actor claimed access to 1.6 million patient records, and multiple class actions were filed. The company’s ad network included fabricated physician personas, “Professor Albust Dongledore,” “Dr. Tuckr Carlzyn MD,” running over 5,000 Meta ads alongside a website disclaimer that these individuals “may be actors or AI portraying doctors.” The Times told the story of a man who used AI to build a billion-dollar company alone. The article was really a transcript of a Silicon Valley fever dream. A byproduct of regulatory lag and consumer desperation A billion dollars in pharmaceutical transactions running through an organization with no one seeming to care if a product can survive contact with the human stomach better than a chicken nugget.

Exciting session on AI in medicine, industry-academic partnerships. Insights from leaders: @euanashley, @james_y_zou, @rustyhofmannMD @Stanford, Robin Lorenz @genentech, @aametwally1 @Google, @jackpo @ Ansible Health. AI+biology will redefine discovery, clinical care.

A small preview of how we scale care for tens of thousands of patients and the rigor behind how we build these tools :)

@BeyondOrigin @AnthropicAI it's not real, I think they broke something. I'm still limited by weekly token limits in my account that had tripped the wire.

@AnthropicAI the entire usage section was remade? No more weekly usage? no more sonnet usage? We have monthly usage now? we are lost here guys

@JeffDean my heart breaks for this ICU nurse who died literally helping people.

This is absolutely shameful. Agents of a federal agency unnecessarily escalating, and then executing a defenseless citizen whose offense appears to be using his cell phone camera. Every person regardless of political affiliation should be denouncing this.

@wizaj @AmadeusITGroup @amadeus4dev did you get a reply from them? Did you find another way to get what you need? I'm looking for something similar.

May anyone here connect me to the API team @AmadeusITGroup @amadeus4dev? Struggling to get a prod API key for my open source flight search CLI which is designed to be super easy for AI agents to use: github.com/wizaj/flight-b… The go-live on their portal has been down for days.

@amadeus4dev are you guys still supporting self-service APIs? My prod request has been pending for 3 weeks now, what's going on over there?

I turned chess into a video game: 40 players. Giant chessboard. ONE winner. Also, I added bombs💣 Playtest starts Oct 25th!

.@AcademyHealth joined the Friends of AHRQ in sending a letter signed on by 194 organizations urging Congress to fully fund AHRQ in FY26 and beyond. Read the letter and share with your networks here: academyhealth.org/blog/2025-03/p…

Omniparser-v2 is ridiculously slow and error prone to me (using GPT4o / o3-mini). Operator seems miles ahead. Has anyone else also set it up and found a different experience?

The Republicans’ Choice: Tax Cuts for the Rich or Health Care for Americans nytimes.com/2025/02/02/opi…

@Azure @github the azure deepseek model just outputs garbage? What's gong on? Is it incorrectly deployed?

DeepSeek R1 is now live on Azure AI Foundry and @GitHub. Experience the power of advanced reasoning on a trusted, scalable AI platform with minimal infrastructure investment. Learn more: msft.it/6016UVP9a #AzureAIFoundry

“OpenAI huh? $500B? Gimme 2 weeks and $5 million and I’ll clone it asap” - DeepSeek’s CTO

This deepseek shock from those who pretend to be in AI pretty hilarious. Those of us actually doing AI have always known China was neck and neck for years now.... Deepseek R1 also came out a MONTH ago and all of us have been using it.. why the shock now?

i'm comically impressed that people are coping on deepseek by spewing bizarre conspiracy theories -- despite deepseek open-sourcing and writing some of the most detail oriented papers ever. read. replicate. compete. don't be salty, just makes you look incompetent.

@deedydas I don't think anyone ie saying China is crushing US. On the other hand, this false belief that China is years behind is also crazy. Realistically anyone who actually works in AI realizes China has always been basically neck to neck...

The China is crushing US rhetoric totally forgets about Gemini 2.0 Flash Thinking. Likely cheaper, longer context and better on reasoning. We're still early in the AI race.

@AnilMakam I think you might be biased by UCSF, I have not seen starting NP command 80% to 90% of a freshly minted hospitalist physician salary.... And all of our NPS work weekends / holidays / sick calls.....

A starting NP can command 80-90% of a freshly minted hospitalist physician salary but care for many fewer patients, doesn't work weekends/holidays or sick call, capped at 40 hours, and is supervised by an attending hospital medicine, so minimal liability Is this high value?