Tom Stanis

@cwhogg i kind of feel sorry for these guys

@tomstanis Did you know it takes 17 years for discoveries to make it into practice? 🤯

Chief Officer of Longevity would have been cooler.

fun facebook.com/reel/133924304…

@dp_oneill Clearly we need to cut all this innovation out and get back to good old care denying as the differentiator.

One promising way to save taxpayer money and protect seniors would be to slash these supplemental benefits. These things are mostly just taxpayer-funded marketing gimmicks that lure seniors into plans that otherwise don’t make sense, and inflate revenues for insurers & brokers.

@dvasishtha There are non nostril pillows as well

The sleep therapist I saw (shout out to her) told me she always offers patients whose specialist ask for masks with pillows one without them. Turns out pillows improve air pressure delivery but are uncomfortable. Both in literature and anecdotally they drive lower adherence.

Banal reason of the day healthcare is broken: - PCPs are referring more chronic conditions to specialists (cardiologists, ENTs, neurologists, pulmonologists) - specialists suck at managing chronic conditions Recent example from picking up my CPAP machine:

@dp_oneill @cwhogg @StuartBlitz Diagnosis codes is my guess.

@cwhogg @StuartBlitz My question would be: What do they mean by "encounter criteria in the claim?" Typically, the differences between 99214 and 99215, say, could only be gleaned by seeing the actual visit note, no? How long was the visit, was there Rx management, etc.

I feel like the risk of consumers changing their healthcare cash spending habits is lower than the risk of payors making it harder to bill/reducing payments? The latter seems inevitable.

@StuartBlitz We can achieve both good access AND better relationships if we are clever.

I think for sure there are examples of medical care becoming too transactional over the last few years, but IMHO medical care has historically also been too paternalistic and too hard to access. Like most things, finding the appropriate balance is the optimal solution.

This is going to help so many patients.

Pollen Robotics, a Hugging Face subsidiary, has open-sourced Amazing Hand, a 3D-printable, four-finger robotic hand with eight degrees of freedom (DoF). Weighing 400g and costing less than $250 - it is designed for Reachy2’s wrist but adaptable to other robots.

@nikillinit We did this in early days of Google. It was called Prophet. You could also short objectives.

here's my new business idea give everyone at your company $10 a month to bet on internal, company-specific prediction markets "will this company turn into a customer" "how much revenue will we hit this year" "will we succeed in this pilot" use it as information mining of what your employees actually think about internal initiatives

@cwhogg bolt, replit, lovable are the hot ones... not sure the best.

Lazy twitter: what are the best AI website generation tools? Not content for the site, but the site itself.

@chrissyfarr Then they would be RSUs

Options should be self executing and never expire!

Still boggles my mind that we make startup employees pay taxes on their shares in a company while it’s still Monopoly money (RSUs) OR require them to buy their “options” — a stake they’ve already earned by working there — after leaving the company. So unfair for employees.

@StuartBlitz What if I redline the methods of destruction?

If you take more than 1 minute to review and sign an NDA, you’re ngmi

That's a good validation of the problem. Having the LLM write the code and call tools seems reasonable as that is how a human would do it... but I'm not sure I trust the LLM to write the correct code for general prompts. Another approach I've considered is to break up the record into documents, have the LLM classify the documents, and build a knowledge graph. This is another way of solving the context limit problem. You can think of this as another version of the chain of thought pattern.

This is a pivotal question. The short answer is that my FHIR MCP tools are firmly in "PoC" land... but to show at least one way you could create an eval (inspired by my analysis this afternoon) consider a form of "needle in the haystack" task where you give the model a big pile of JSON and ask it to find the average blood pressure. I "wrote" (had gemini write me) a little script to mix my blood pressures in with a pile of distractor resources from @MITREcorp's Synthetic Mass project (filtering out all BPs from that set), and then I gave the pile of JSON to a fresh gemini session. Attempt 1: 5000 distractors + 8 real BPs. Result: >1M tokesn, too large for Gemini contex. Attempt 2: 1000 distractors + 8 real BPs. Result: 640K tokens, Gemini found and averaged 5 of the BPs (but missed 3). Session at: aistudio.google.com/app/prompts?st…

Follow-up to the idea of LLM agents + tools for EHR workflows: Here's a practical demo showing it connect to real EHR data. An MCP server offers tools (e.g., grep, eval_js) for the LLM to analyze the EHR, moving beyond just context stuffing. See it query conditions, notes & vitals. Full post + video in reply...

@JoshCMandel This is super cool. Josh, do you have any evals that you have built or thought about around this? I'm curious about your comment that "LLMs need tools to deal with this data". Would be great to see an eval to see where the tools can boost performance on tasks.

Post: linkedin.com/pulse/theory-p… Video: youtu.be/K0t6MRyIqZU

I’ve had weekends where I felt completely lost, only to show up Monday with a plan I was excited to share. Not because I found the answer, but because I stopped trying to solve and started trying to describe. Here’s how I use GenAI to break out of mental loops and rediscover momentum: @tstanis/out-of-the-fog-how-ai-helps-break-agency-traps-288dc5a5e262" target="_blank" rel="nofollow noopener">medium.com/@tstanis/out-o…

@drdanchoi @mcuban While I appreciate the sentiment, we need to have credible numbers. $25k is nowhere near the cost of a heart transplant. Add a zero and then some.

.@mcuban asked “Are doctors overpaid?” Credit: How I Doctor Podcast with Dr Graham Walker

@StuartBlitz The pricing model should be $0.10 per page. Good deal for all. Good margins.

Making my next angel investment. Talked to a physician friend who had a unique insight that medical data is still being faxed and shouldn't be in 2025. He is creating a HIPAA compliant e-mail service so you can just e-mail medical data to anyone with 1 click. Anyone else in?