Felix Hofmann

Surgeon @ LMU Hospital Munich | MD, Dr. med., | special interest in HPB, real world data, synthetic patients & AI applications

[Yohei @yoheinakajima]

needing a technical cofounder was a big blocker for industry experts who wanted to solve a problem they noticed by building a startup

excited to see more industry experts get started without needing a tech cofounder, even if they need to hire real devs once launched

[Fabian Theis @fabian_theis]

Munich takes the top two spots in the FT ranking of Europe’s startup hubs — UnternehmerTUM #1, Start2 Group #2.

Strong science, engineering depth, & real bridges between research + entrepreneurship. Huge opportunity ahead in part. for AI + bio/med. 🚀
https://t.co/cTJswhhJfr

[Michał Podlewski @trajektoriePL]

Cardiologist wins 3rd place at Anthropic's hackathon. Out of 13,000 applications. Built in 7 days by Michał Nedoszytko MD. Coded day and night - in the hospital, in the cloud, while flying from Brussels to San Francisco.
A few years ago, it would have been impossible for a doctor to build this alone in just a couple of days. AI changed that.

The project is called https://t.co/wAliajqjVF. It is an AI agentic care platform for patients. Including reverse AI scribe it is a companion that guides the patient from the moment they leave the doctor's office.

Powered by the massive context window of Opus 4.6, it allows patients to explore their full medical history, connected devices, Evidence Based resources and external data sources — all in one place.

Today, the barrier to entry has vanished; even a practicing physician can build an application from scratch.

[@levelsio @levelsio]

🇪🇺 As a European citizen and AI founder, I can apparently use these "AI Factories", so I just signed up to use them!

Every "supercomputer" has an [ ACCESS NOW ] button which made me very excited

I expected to sign up, maybe pay a discounted H100 rate (funded by EU, that'd be nice?) and get a Jypyter notebook, or some SSH login so I can access my GPU like I'd do on @lambdaapi or @awscloud or @Hetzner_Online

But I celebrated to early, I signed up, confirmed my email, then ended up in a "Supercomputer Access Calls" page, where I had to select from a tedious list of "Call For Proposals" to get access to a GPU

So I could NOT just access a H100 GPU, I have to make sure my project (in this case my business) fits a specific proposal, ok fair

This process was already tedious enough but then when I tried to actually go through with it, it started asking me if I had "Respect for Human Agency?", I do I think, and if I was mindful of "Individual, and Social and Environmental Well-Being?", well I am, right guys??? Right??? The questions didn't stop, just endless pages of this

Look I get what they're doing, they pivoted the classic university "I need to rent a giant computer for my research" to an EU wide thing and then present it as the "European AI plan"

But this isn't really how AI works in production? As a founder in AI, if I wanna do stuff I'd rent a whole bunch H100 GPUs again at @lambdaapi or @awscloud or @Hetzner_Online and SSH into a box

Or if I want it more simple I run AI models on @FAL, @wavespeed or @replicate which is just an API call or web front end I can click stuff and run a model

The EU has the right intentions here but it's just the wrong execution, this thing will 100% go nowhere, and I'm a born optimist, I want to believe, I'm also a proud European, and I'm in AI a bit and not a complete idiot. There's just better ways to do this

If you really want to have the GPU servers in Europe (which arguably isn't that important), then let me rent a GPU box with SSH access at @Hetzner_Online or @OVHcloud that's hosted in Europe and subsidize that for European citizens and European businesses. I don't even believe in that, but at least that'd make it accessible for Europeans. Now it really isn't?

What's REALLY much more important though if you want to be a part of the AI race and I've posted for years here with @euaccofficial is to make Europe a really extremely attractive place to start and run an AI business. Remove regulatory obstructions and give tax discounts for startups. Let them build a business first that can compete worldwide and once they make enough money (let's say $100M/y), then slowly start adding regulation. Because right now the regulation only benefits the European incumbents, the dinosaur companies, while making it very difficult for European citizens to start new AI companies here.

Which is why we literally have none left.

Anyway, I applied to get my GPU, let's see if I get it!

[Thomas Stoop @TFStoop]

Published in @TheLancet, our review on #PancreaticCancer
📎Epidemiology🌎
📎Pathogenesis🔬
📎Risk factors & surveillance🕵️
📎Diagnostics🩺
📎Treatment💊
📎Future directions🔍
… by @ammar_asrar, @ObaAtsushi, T. Seufferlein, JW Wilmink, @BasGrootKoerkam & @MarcBesselink

[Artificial Analysis @ArtificialAnlys]

DeepSeek takes the lead: DeepSeek V3-0324 is now the highest scoring non-reasoning model

This is the first time an open weights model is the leading non-reasoning model, a milestone for open source.

DeepSeek V3-0324 has jumped forward 7 points in Artificial Analysis Intelligence Index, now sitting ahead of all other non-reasoning models. It sits behind DeepSeek’s own R1 in Intelligence Index, as well as other reasoning models from OpenAI, Anthropic and Alibaba, but this does not take away from the impressiveness of this accomplishment. Non-reasoning models answer immediately without taking time to ‘think’, making them useful in latency-sensitive use cases.

Three months ago, DeepSeek released V3 and we we wrote that there is a new leader in open source AI - noting that V3 came close to leading proprietary models from Anthropic and Google but did not surpass them.

Today, DeepSeek are not just releasing the best open source model - DeepSeek are now driving the frontier of non-reasoning open weights models, eclipsing all proprietary non-reasoning models, including Gemini 2.0 Pro, Claude 3.7 Sonnet and Llama 3.3 70B. This release is arguably even more impressive than R1 - and potentially indicates that R2 is going to be another significant leap forward.

Most other details are identical to the December 2024 version of DeepSeek V3, including:
➤ Context window: 128k (limited to 64k on DeepSeek’s first-party API)
➤ Total parameters: 671B (requires >700GB of GPU memory to run in native FP8 precision - still not something you can run at home!)
➤ Active parameters: 37B
➤ Native FP8 precision
➤Text only - no multimodal inputs or outputs
➤ MIT License

[Nicholas Hornstein @GIMedOnc]

#GI25
Is it the hardest to pronounce name of any antibody? Yes.

It is also probably the best new treatment for GI malignancies? Also, probably yes.

Daraxonrasib, pan-KRASi
Amazing responses in second line PDAC
Making the undruggable druggable.

(Someone please teach me how to pronounce this.)

[andrew chen @andrewchen]

Bureaucrat mode:

- create committees for every decision
- make sure every meeting has pre-meetings to build the papers/deck for the meeting
- end every meeting by expanding the scope of the project
- no one ever owns a decision, so make sure there's complete consensus. Create more meetings if needed
- punish anyone showing initiative
- discipline anyone who moves quickly without complete consensus
- require detailed status reports before any progress
- create complex approval workflows for trivial tasks
- celebrate vanity metrics and milestones
- reward people based on "impact" based on how may people are working on your projects
- ask legal, brand, compliance to approve everything no mattery how small
- talk endlessly about downside risk

... what else?

[Bent Freiwald @BentFreiwald]

Wir sollten junge Menschen wie eine Minderheit behandeln.

Aus einem einfachen Grund: Sie sind eine. Das schreibt selbst ein Expertengremium der Bundesregierung. Und nennt drei Gründe, die ihr kennen solltet. THREAD 🧵

[Thomas Wolf @Thom_Wolf]

Strong picture!

[Matthias R. Pfister @Pfister_M_R]

📣📣 Finally online - #EuropeanSchoolOfSurgery 📣📣

Freely accessible, global, online learning platform!

More than 120 Videos 📽️ free available, check it out:

🫵🏻https://t.co/0PoGN07F1R

[Science Magazine @ScienceMagazine]

A new ingestible robotic device electrically stimulates the intestines to trigger the resumption of peristalsis during a form of intestinal paralysis called postoperative ileus.

Learn more in @SciRobotics: https://t.co/FlXm7vu3zl

[Daniel Donoho, MD @ddonoho]

This is a great example of everything that is right, and everything that is completely broken, in medical “AI” research

If we keep doing *this* we will never get *there* …

Warning 🌶️🌶️ takes ahead

What do I mean?

This is great! Every pediatrician could use this app on every ear exam … it might* make some* of them better* …

Yes, 1/4 of kids had too much earwax

No, we don’t know how this compares to the clinical data (ie were the kids treated by their pediatrician as AOM or not) presumably because getting that data was too hard/messy.

but that’s not the real problem

Building *the next thing* is no closer than it was before this article was published…
1. Took FIVE YEARS to accrue data from two sites.
2. No availability of data models or code. “Data available to researchers with a data sharing agreement (supplement 2” = send me an email, wait 3-6 months, and then maybe.
3. Proprietary, commercialize/IP then publish, single purpose model on 2017 era architecture (sorry…) that doesn’t scale to any other use case
4. Even the article is paywalled ( like most journals)
5. We already knew that this could be done (multiple prior papers on this exact question)

We have to recognize where we are, applaud the authors for their good work, and not accept this state of affairs.

[Andrew Soltan @andrewsoltan]

Training fairer medical AI needs diverse data, but hospitals are restricted in data they can share for privacy reasons.

We built an easy-to-deploy platform for hospitals to take part in AI development without sharing data, and piloted it at 4 NHS Trusts

https://t.co/VA41fXutdJ

[Felix Hofmann @FelixOHofmann]

@ZhiVenFongMD @Mon_ish_K FTR (=deaths/complications = mortality/morbidity = „compl. specific mortality“ but fancy?) never accounts for volume, does it? I can see utility in exploring well defined questions (e.g. mortality after PPH, management A vs B), but as a measure (not concept), in general?! idk…

[Felix Hofmann @FelixOHofmann]

life (or disease or death) can be seen as a consequence of events. @suneman et al using transformers to capture people's life trajectories is a nice example, how embeddings make these high-dimensional information available for downstream analyses. great paper!

[Felix Hofmann @FelixOHofmann]

also discussed by @JorgKleeff and U. Ronellenfitsch https://t.co/WqyAkOnqpA

[JorgKleeff @JorgKleeff]

AI and imaging-based cancer screening: getting ready for prime time https://t.co/sFXJCIiY48
https://t.co/fCQ4ab8gl1

[Felix Hofmann @FelixOHofmann]

In non-contrast CT scans, a deep learning model (nnU-Net based once again ☺️) was better in differentiating pancreatic lesions than the average radiologist (https://t.co/PLq7CFM8tu). Should we use AI to screen for pancreatic lesions in patients who underwent CT for other reasons?

[Ville Sallinen @villesallinen]

Federation of Finnish Learned Societies (most important journal ranker in 🇫🇮) proposes all MDPI, Frontiers and Hindawi journals to ranked as class 0 due to questionable behaviour.

[Ville Sallinen @villesallinen]

Oh yes, PORSCH trial with 50% reduction in mortality with algorith-based postop care after pancreatic resection @TheLancet

https://t.co/ttPICpq5Rh
@MarcBesselink