Krzysztof Karwat

MD • Interventional Cardiologist • Training diffusion models after hours

[Krzysztof Karwat @KPKarwat]

Ok, here is the next iteration of our model. This kind of plaque ruptures in 30% of simulations.
#OCT #Atherosclerosis

[Krzysztof Karwat @KPKarwat]

If you need a 3D model of a TAVI valve, just generate it procedurally with Claude Opus 4.7.
Check out the transformation across these four distinct iterations.
#ClaudeAI #TAVI #TAVR #3DModeling

[Krzysztof Karwat @KPKarwat]

@blueeyedboy88 Warto wypróbować https://t.co/WaxFzSvdcV wytrenowane na 2M ekg

[Krzysztof Karwat @KPKarwat]

Aggregating all medical data into a single general model could lead to cross-modal confusion, such as retrieving CT signals from MRI images, especially in lower-quality images. From a clinical standpoint, each modality presents distinct constraints and artifacts that doctors are trained to filter out.
Keep in mind that the current models always give an answer, even for data points out of the training distribution. This is not the way clinicians work. But hopefully, the MoE model approach, utilizing a specialized expert for each modality, might be the solution

[Krzysztof Karwat @KPKarwat]

@SportsdreamW @callmeMizuko No

[Krzysztof Karwat @KPKarwat]

@MainatJM @Mishi_2210 Better ¹¹5118

[Krzysztof Karwat @KPKarwat]

@MainatJM @Mishi_2210 Tetration 🙂 ¹¹505

[Mariusz Kurman @mkurman88]

This is it: A single-person project; Trained from scratch on TPUs (@Google TRC) on the one and only SYTNH dataset by @pleiasfr;

Neuroblast-v3 architecture running on my local vLLM instance

Just wow (I'm amazed by how good it looks; speed is incredible, here slightly slowed by high-resolution recording)

Todos
> needs agentic fine-tuning in the future
> needs some fine-grained RL

[Krzysztof Karwat @KPKarwat]

@mkurman88 Genuine AGI vibes 😁

[Krzysztof Karwat @KPKarwat]

@KinasRemek Chcę zastąpić VAE przez triplane/nerf na obrazach medycznych i wytrenować na tym model dyfuzyjny. Na pewno będzie wesoło 🤭

[Krzysztof Karwat @KPKarwat]

@mkurman88 Yes, he will establish a new dynasty

[Krzysztof Karwat @KPKarwat]

@DrJayMohan @ESHLOF @AntoniousAttall @AbbottCardio @Babar_Basir @adityadoc1 @SVRaoMD @djc795 @rajivxgulati @SandeepNathanMD @nadia_sutton Eruptive: casts shadow on OCT

Noneruptive: no shadow

[JAMA @JAMA_current]

A bioresorbable occluder offers a promising alternative to metallic devices by effectively sealing an atrial septal defect, promoting endothelialization, and gradually degrading to leave native functional tissue.

📽️ Watch the video for a demonstration of the design and implementation of a bioresorbable occlude.

📖 Read the full RCT: https://t.co/OiszNCmaPf

[Krzysztof Karwat @KPKarwat]

@TrackYourHeart

[Andrew Ng @AndrewYNg]

Really proud of the DeepLearningAI team. When Cloudflare went down, our engineers used AI coding to quickly implement a clone of basic Cloudflare capabilities to run our site on. So we came back up long before even major websites!

[Michal Domanski @moustache_doman]

Złożyłem podpis pod listem otwartym środowiska Open Source w sprawie cyfrowej niezależności.

Zarówno w imieniu prywatnym, jak i Bielik Consulting.

I nie, nie chodzi o psioczenie na Bóg Techy.

Chodzi o pokazanie, że w Europie jest potencjał, z którego można korzystać.

O to, że państwa europejskie, realizując swoją działalność, mogą korzystać z europejskich technologii i talentu.

Sektor publiczny ma alternatywę.
Pozostaje mieć jeszcze wolę i chęci.
Oraz system wspierający taki stan rzeczy.

My, jako gracze w tym ekosystemie, jesteśmy gotowi. Mamy wokół siebie wspaniałą sieć dostawców technologii, sprzętu, know-how.

My nie tylko jesteśmy w stanie wspierać sektor publiczny na światowym poziomie - już to robimy. A możemy robić więcej.

_____________
Full text 👇

On the occasion of the French-German Summit on European Digital Sovereignty on 18 November 2025, the European Open Source industry associations and their members have come together to declare that digital independence is possible.

Digital Sovereignty for Europe means that we take back control over digital technology – from its design and development, to daily operation and long-term maintenance – creating resilience against economic pressure, attacks on democratic sovereignty, or even military aggression, and avoiding lock-ins to proprietary ecosystems.

Europe is an important market for international businesses, but more so, Europe has the capacity and talent in research and development, business, and finance to shape digital technology respect­ing our common values and contributing to welfare and prosperity.

In order to seize the opportunity and unlock the full potential for the European economy, the un­dersigned European Open Source industry associations and their member companies provide products, services and solutions that implement alternatives for the government stack and put an end to digital dependencies in the public and the private sector.

We offer the full stack of enterprise-grade solutions from network infrastructure, data centres and cloud infrastructure, operating systems, software development tools and build pipelines, to office desktop solutions, cooperation and communication apps, and much more.

We offer the expertise, services, and local competent partners to realise the digital sovereignty goals of France, Germany, and the European Union.

On the occasion of the French-German Summit on European Digital Sovereignty, we are declaring that digital independence is possible, and call on the President of the French Republic, and the Federal Chancellor of the Federal Republic of Germany to make use of the resources we, the industry en­abling Open Source, provide to promote and realise digital sovereignty, and as an overall lever for European global competitiveness, economic sovereignty, and innovation.

[Krzysztof Karwat @KPKarwat]

@mkurman88 Nice, 60M model could run even on electronic thermometer 🙂

[Krzysztof Karwat @KPKarwat]

@snezanalawrence @PhysInHistory Kraków

[Krzysztof Karwat @KPKarwat]

@illscience Betting on No. 2 with medical AI and European Union loosening GDPR law for training models on medical data