Country First - America is Louder Virtual Campfire x.com/i/broadcasts/1…

We have a very poor understanding of why deep neural networks like transformer models learn the parameters they learn. For example, in the paper below from 2013, the authors demonstrated that 5% of the weights of a trained deep neural network can be used to predict the values of

The in-context learner of the "beautiful Google Research paper" is a meta learner like Sepp Hochreiter's 2001 meta LSTM [1] which learned by gradient descent (GD) a learning algorithm that outperformed GD - no test time weight changes! Since 1992, GD can learn learning algorithms

👏 Congrats to MichaelNg, Nikhil Buduma, the entire Ambience Healthcare team — rare good guys in my "50 Most Popular AI Scribe" ranking: sergeiai.substack.com/p/top-50-ai-sc…. Just a couple of years ago, they were scrambling. But I’ve gotta say, they’ve drastically improved and seriously polished their

🔥 Today is a massive podcast drop. “Is There a Science Behind Longevity?” — a new episode I co-hosted with my good friend and digital health expert AlexKoshykov — just dropped on Digital Health Inside Out: youtube.com/watch?v=QSKIIp…. Guests include such renowned experts as

I went to Europe and found the longevity cure. 😉 It’s called a “fuck it” attitude. Americans should try it. They’d probably live longer. In my obsessively detailed review, I go beyond facts and data to back that up, plus address 35 other popular questions and myths about

The White House’s “preferred” list of 63 “Make Health Tech Great Again” companies fails to represent the industry I know. (The link to the article is in my bio.) I’ve never heard of most of these companies, especially on the "apps" side. And this is literally my job: to know

Who invented convolutional neural networks (CNNs)? 1969: Fukushima had CNN-relevant ReLUs [2]. 1979: Fukushima had the basic CNN architecture with convolution layers and downsampling layers [1]. Compute was 100 x more costly than in 1989, and a billion x more costly than

I'm attending HLTH USA 2025, Oct 19-22 in Las Vegas! Join me at healthcare's #1 must-attend innovation event app.ingo.me/q/5h3io #HLTHUSA

All of the peasants relaxing while the important people are working 💪🏻 🖥️

🚨 I’m honored to be speaking at the Digital Health & AI Innovation Summit (DHAI) in Boston 🏙️ on June 8–9, 2026 📅. Grateful for the chance to join this amazing community shaping the future of healthcare 💡🩺. Hope to see many of you there 🙌 worldbigroup.com/digital-health…

OpenAI is pushing #HealthBench in every model release, including #gptoss today, like it's a clinical breakthrough, when it's really just a benchmarking gimmick.

Was waiting for this, Sergei Polevikov and Sergei weighs in with his usual delicate and polite manner on the latest from OpenAI. #AIinHealthcare sergeiai.substack.com/p/openais-gpt-…

Dr. Katalin Karikó who won a Nobel Prize for developing mRNA vaccines responds to RFK Jr: "When you are sick, you wouldn't ask for advice from a lawyer like Kennedy. You ask experts, your physician, for advice."

Calc Consulting Maybe OpenAI actually does know its layers after all 😉

➤ OpenAI just released its first open-weight language models since GPT-2, called #gptoss. ➤ They keep pushing HealthBench like it’s proof they’ve “solved” healthcare. 🙄 ➤ You can’t really fine tune gpt-oss. So unless your dataset looks like #HealthBench, you’re out of luck.

🚑✨ New Interview Drop! ✨🚑 Ever wondered how a major health system REALLY decides which startups get a pilot—and which ones get shown the door? I sat down with Komal Bajaj MD, MS-HPEd Bajaj, CQO at NYC Health + Hospitals, to talk: youtu.be/fPqj6ECa55Q #DigitalHealth #HealthTech

There never existed a more despicable nation than russians

𝗔𝗿𝗲 𝘁𝗵𝗲 𝗢𝗽𝗲𝗻𝗔𝗜 𝗢𝗽𝗲𝗻𝗦𝗼𝘂𝗿𝗰𝗲 𝗺𝗼𝗱𝗲𝗹𝘀 𝗼𝘃𝗲𝗿𝗳𝗶𝘁 𝘁𝗼 𝘁𝗵𝗲𝗶𝗿 𝘁𝗿𝗮𝗶𝗻𝗶𝗻𝗴 𝗱𝗮𝘁𝗮 ? Analysis with weightwatcher shows over 10% of the layers in the 'red zone', and, oddly, closer to the labels than the data. It is reminiscent of the Llama