Ran Shaul

Co founder, Chief Product Officer @khealth

[Mark Tluszcz @marktluszcz]

Specialised players like @khealth are better positioned. They are working with large Healthcare Systems and running their AI suite of Agents in actual clinal settings…in April they published a relevant study with @CedarsSinai relayed by @Forbes in this article: https://t.co/D6v5LfA1b2

[allon bloch @allonbloch]

Mass General Brigham and @khealth can fix primary care using AI and new approaches to scaling high quality care delivery in MA and beyond
https://t.co/ZHQZcTojSD

[allon bloch @allonbloch]

We're big fans of @CMSGov work is doing around freeing up data and leveraging AI to empower patients and doctors. Led by @DrOz and the A team of Amy Gleason, Arda Kara and Alberto Colon
Yesterday we pledged to support this initiative

https://t.co/ctG9aAMbnW

[allon bloch @allonbloch]

Exciting to see our growing partnership with @MayoPlatform with many health systems that will be adopting @khealth AI Physician Mode capabilities

https://t.co/lzp9j4GdOV

[Dan Zeltzer @danzeltzer]

🚨New @AnnalsofIM paper: In 461 @CedarsSinaivirtual urgent care cases with common symptoms, @khealth AI recommendations were rated optimal more often than physicians' (77% vs 67%), & less often potentially harmful (2.8% vs 4.6%). https://t.co/KnIptbAdtI

[Eric Topol @EricTopol]

When physicians with access to A.I. were compared with A.I. for virtual urgent care visits
The A.I. performed better.
@AnnalsofIM
https://t.co/PAh75ckpHd

[allon bloch @allonbloch]

We are announcing today the results of a major study of K Health’s AI Physician Mode study with Cedars Sinai. Published in the Annals of Internal Medicine.

The bottom line is K’s AI has performed better than doctors in a whole host of acute complaints in real life primary care settings.

This is based on 8 years of building our platform:

real high quality data
real AI co-pilot in medical care delivery
real patient settings

This is not a medical textbook comparison. These are real life patients.

Study: https://t.co/dEgbRPZjdL

[allon bloch @allonbloch]

Bethany I hope you're doing well! 
After reading your scary medical story I want to point out that while you think Chatgpt saved your life, in fact it gave you bad medical advice on the outset and missed a dangerous escalation in your health.

Simply put, a petechial rash is a red flag warning sign that should be immediately addressed in an ER setting, but instead it accepted the suggestion that the cause was dietary. That is a huge miss and could have dangerously delayed your care. ChatGPT is an amazing translator of medical terms but is not a full blown system of care. more in the thread below

[Gavin Baker @GavinSBaker]

1) DeepSeek r1 is real with important nuances. Most important is the fact that r1 is so much cheaper and more efficient to inference than o1, not from the $6m training figure. r1 costs 93% less to *use* than o1 per each API, can be run locally on a high end work station and does not seem to have hit any rate limits which is wild. Simple math is that every 1b active parameters requires 1 gb of RAM in FP8, so r1 requires 37 gb of RAM. Batching massively lowers costs and more compute increases tokens/second so still advantages to inference in the cloud. Would also note that there are true geopolitical dynamics at play here and I don’t think it is a coincidence that this came out right after “Stargate.” RIP, $500 billion - we hardly even knew you.

Real: 1) It is/was the #1 download in the relevant App Store category. Obviously ahead of ChatGPT; something neither Gemini nor Claude was able to accomplish. 2) It is comparable to o1 from a quality perspective although lags o3. 3) There were real algorithmic breakthroughs that led to it being dramatically more efficient both to train and inference. Training in FP8, MLA and multi-token prediction are significant. 4) It is easy to verify that the r1 training run only cost $6m. While this is literally true, it is also *deeply* misleading. 5) Even their hardware architecture is novel and I will note that they use PCI-Express for scale up.

Nuance: 1) The $6m does not include “costs associated with prior research and ablation experiments on architectures, algorithms and data” per the technical paper. “Other than that Mrs. Lincoln, how was the play?” This means that it is possible to train an r1 quality model with a $6m run *if* a lab has already spent hundreds of millions of dollars on prior research and has access to much larger clusters. Deepseek obviously has way more than 2048 H800s; one of their earlier papers referenced a cluster of 10k A100s. An equivalently smart team can’t just spin up a 2000 GPU cluster and train r1 from scratch with $6m. Roughly 20% of Nvidia’s revenue goes through Singapore. 20% of Nvidia’s GPUs are probably not in Singapore despite their best efforts. 2) There was a lot of distillation - i.e. it is unlikely they could have trained this without unhindered access to GPT-4o and o1. As @altcap pointed out to me yesterday, kinda funny to restrict access to leading edge GPUs and not do anything about China’s ability to distill leading edge American models - obviously defeats the purpose of the export restrictions. Why buy the cow when you can get the milk for free?

[Ran Shaul @ShaulRan]

Open-source models empower application layer developers to construct precise and highly customized models tailored to specific use cases. This ensures that the competitive edge lies not in foundational models themselves, but in creating value directly for the end-user

[Mark Tluszcz @marktluszcz]

Impressive results from @deepseek_ai & it's open source...could this be the "coup de grace" for LLMs?

Have they really built it on the cheap? I hope so...yet they might have had some help:

https://t.co/C7sOt13o2p

[Ran Shaul @ShaulRan]

@GavinSBaker Thank you for making sense!

[Ran Shaul @ShaulRan]

Interesting perspective. I do see a slowdown in the modeling capabilities. There's a lot happening on the application layer, but not so much in the modeling capabilities themselves

https://t.co/mDTLwWJCRV

[Ran Shaul @ShaulRan]

It feels right. It's hard to imagine a future where a physician wouldn't team up with an AI physician, much like a pilot uses autopilot. Both are highly skilled professions.

[Peter H. Diamandis, MD @PeterDiamandis]

I think we have less than 5 years until it will be considered malpractice for doctors to diagnose patients without AI.

[Mark Tluszcz @marktluszcz]

This seems about right from @Benioff "LLMs upper limit reached…future is AI agents"

In the healthcare industry this is 💯 true as seen by @khealth - the AI physician is coming to a practice near you….

https://t.co/vi3ElaeMN6

[allon bloch @allonbloch]

@marceloclaure and @GraciasAntonio speaking about @khealth @FIIKSA
They share our journey to transform primary care through clinical AI, solving the need to provide high quality access to medicine for all humanity

[allon bloch @allonbloch]

Super excited to announce @khealth major new partnership is with @HMHNewJersey (Hackensack Meridian Health). Together with @BobCGarrett Dr Dan Varga, Jim Blazar and their teams we will significantly increase access to high quality medicine in NJ. @ShaulRan
https://t.co/HsCKza7xia

[allon bloch @allonbloch]

We've built the first LLM Agent that's grounded in the patient's EMR that can provide personalized answers for clinical and health related questions.

https://t.co/3Yme7frF9I

[allon bloch @allonbloch]

United Optum telehealth and Walmart Health closing mean that we're finally moving to primary care that works - in @khealth case it's 24X7 clinical AI that is embedded into major health systems and allows personalized care and true LPR
https://t.co/qWzWwouSnk
https://t.co/Rtlem9AHqG

[adam singolda @AdamSingolda]

Would you let your kids go to @Columbia ?

I feel for the parents who worked their entire life, who took loans, mortgaged their house, “pursuing the American dream” of — one day sending their kids to a great school, to get great education.

And then one day wake up and go “honey, isn’t that …”.

Seeing their kids on the @Nytimes on the wrong side… (of history).

Talk about Ivy League going Oopsi …

https://t.co/AAe5GjJKO1

[adam singolda @AdamSingolda]

Scott ... Israel loves you. You use numbers, and data to let the story tell itself.

Here is the thing. This war will end, we will win, we will get stronger, and in fact we will all move on, like people tend to do.

But one thing will stay ... the pictures, the names, the identity of the wolves around us. They were always there, but we now see them, we know who they are, we see their faces on NYTimes articles, we see them in universities, we see them denying professors from accessing campus, they are protesting, they are out there hating, they are not nice people.

We will win. We will become stronger. But the wolves ... they are here to stay, and at last, we know who they are.

Keep up the good work Scott, we're behind ya.

cc @profgalloway @ShaiDavidai @MSNBC