Jan Clusmann

Clinician Scientist for computational oncology / Resident in Gastroenterology at RWTH Aachen University

[Jan Clusmann @JClusmann]

I love Claude 3.5. But, working in academia, I see some issues with release of new models under previous model's names. This makes clear communication challenging. {Metric} of Claude 3.5 Sonnet was X, {Metric} of Claude 3.5 Sonnet(new) was Y?

You can do better... @alexalbert__

[Alex Albert @alexalbert__]

The new Claude 3.5 Sonnet is one of the best models I've ever used. We listened to the feedback on the old 3.5 Sonnet and worked to improve the new model in a number of ways.

Here are some of my favorite improvements:

[Jan Clusmann @JClusmann]

Exciting work! Very much in line with our work (https://t.co/5FhY9ymyRU). Prompt injection will be a serious threat in every LLM-medical device coming to market. Can't wait to use those devices, but they have to be robust against adversarial attacks.
@luke_lee_ai @jnkath

[(Luke) Donghyun Lee @lukedhlee]

🚨 Multi-agent systems are no longer safe from prompt injection!

In our paper, we introduce Prompt Infection—an infectious prompt injection attack that spreads like a virus across LLM agents, turning your multi-agent system into a network of compromised agents.

TL;DR:
1. One malicious email, PDF, or webpage can steal your data and cost you thousands of dollars.
2. Bigger models ≠ Better security. More powerful LLMs, like GPT-4o, can actually be more vulnerable.
3. Imagine LLM town: a scenario where agents infect each other, leading to significant system failures.
4. We’ve explored solutions to mitigate this threat.

Paper: https://t.co/bMwDP3IQd4
More on threads below 👇

[Jan Clusmann @JClusmann]

I can't wait!

[Jakob Nikolas Kather @jnkath]

Save the date for 12-14 Nov 2025 (next year) in Berlin, Germany: The first @myESMO conference purely on AI and Digital Oncology 🤩 Very happy to organize this together with a large panel of experts. Mark your calendars! https://t.co/beCWIkEYIX
@DKFZ @NCT_UCC_DD @NCT_HD @uniklinik_hd @AACR @ASCO @NatureMedicine @JCO_ASCO @NEJM_AI

[Jan Clusmann @JClusmann]

Listen to the science, and stop subsidizing actions. Federal state of germany funds this with 35.8 billion dollars per year, absolutely insane. Traffic sector is scoring a (very bad) first place. See you in climate court.

[Storch_i @Storch_i]

Der Bund fördert klimaschädliches Verhalten jedes Jahr mit 35,8 Mrd. € (24,8 Mrd. € nur im Verkehrssektor)! Hierdurch werden bis 2030 rund 156 Mio. T CO₂ erzeugt. Das ist doch total irre.

Alle klimaschädlichen Subventionen müssen auf den Prüfstand.

https://t.co/UpbiS3d8hh

[Jan Clusmann @JClusmann]

Increased intake of Manganese - a new way "to love your liver" ?(https://t.co/XAILwBjHOK)

New research from the "Schneiderlab" @C_V_Schneider, led by Simon Schophaus and @ktc_phd published now in Liver International.

https://t.co/ojj9XhXnm8

[Carolin V. Schneider, MD @C_V_Schneider]

Excited to share our aMASLDing research on how machine learning uncovered dietary manganese might protect against hepatic steatosis. #LiverHealth #MASLD (1/10) @ktc_phd @KM_Schneider1 @JClusmann @DrLoomba @tom_luedde
https://t.co/QvSWfkVeoZ

[Jakob Nikolas Kather @jnkath]

New research from @katherlab, led by @JClusmann : "Prompt Injection Attacks on Large Language Models in Oncology" https://t.co/pw6jBb9uJg We show that vision-language models in oncology can be attacked easily, creating malicious output which could harm patients.
@tudresden_de @NCT_UCC_DD @NCT_HD @EKFZdigital @DKFZ

[Jakob Nikolas Kather @jnkath]

New preprint from @katherlab: large language models for end-to-end clincial trial matching. Our system parses study protocols and patient records, and matches them. We evaluate thoroughly against human experts and find >90% correspondence. Led by @Dykex6 🤩 https://t.co/vTnTVd19gq
@NCT_UCC_DD @NCT_HD @DKFZ @DataScienceDKFZ @uniklinik_hd @myESMO CC @tompowles1 @NEJM_AI @Agrawal_Mri @alexia_zz @PepPamies

[Jan Clusmann @JClusmann]

The "Deutschlandticket" is an amazing effort to shift mobility to public transport. It works, it lowers social inequities, and we cannot afford to stop it. Plenty of other ways to save money (and CO2!), see below

[Nurder Koch @NurderK]

Das #Deutschlandticket sollte nicht teurer, sondern günstiger werden. Einsparungspotenzial gebe es genügend.
Aber was weiß ich schon…

[Jan Clusmann @JClusmann]

Thank you for the invitation! Was a great experience leading the AI workshop, together with @AlessioGerussi and @MamathaBhat3 Let's make AIH AI-ready ;)

@jnkath @BastianEngel

[Jan Clusmann @JClusmann]

Medical students in Dresden demonstrating for (at least some) pay during their 6th year internship is something that has to be supported! It reduces quality + equality if MD students have to worry how to pay their rent and buy food, while working 40h + (+ studying!)

[Jan Clusmann @JClusmann]

Great news! https://t.co/eVKKQjcbuG

[Jan Clusmann @JClusmann]

Research in the era of AI session @EASLnews chaired by @C_V_Schneider and @jnkath with amazing talks by Isabella Wiest, Raquel Pérez-López and Stefano Caruso. Amazing talks, and highly relevant discussion. AI literacy is essential, as is scientific methodology / critical thinking

[Jan Clusmann @JClusmann]

Not even CDU-followers want to go back to burning oil for car engines. People are so much more willing to work towards climate neutrality than political parties think...

[Jakob Nikolas Kather @jnkath]

Finally out in @NEJM_AI 🤖, led by @Dykex6 from @katherlab: "GPT-4 for Information Retrieval and Comparison of Medical Oncology Guidelines". We've developed a RAG pipeline using the GPT-4 API to answer medical oncology questions based on guidelines. Previous studies have claimed that "large language models (LLMs) are not suited for oncology" -- but our data debunk these claims! Using RAG, LLMs are very good at medical decision-making in oncology -- as we show in a quantitative evaluation. There is still room for improvement, but LLMs+RAG are the way to go ✨

Paper link: https://t.co/fw0j8OEBg3
Free-access link: https://t.co/GQQjbNCtId

@myESMO @ASCO @AACR @uniklinik_hd @Medizin_TUD @NCT_UCC_DD @NCT_HD @DataScienceDKFZ

[Jan Clusmann @JClusmann]

Important step for our democracy!

https://t.co/DAu6m6KKJN

[Jan Clusmann @JClusmann]

@ArndtVogel @EASLedu @ILCAnews @myESMO This was great! Rarely learned so much (and met so many amazing people) in this short period of time. Huge thanks to @EASLedu and the organizers of the school

[Jan Clusmann @JClusmann]

The science is clear, the law in place since 2019, climate protection is something all democratic parties and Most of DE agree on (to a certain level). It is concerning when a federal minister now creates fear in people instead of taking measures to reduce CO2 in traffic!

[Luisa Neubauer @Luisamneubauer]

Die Bundesregierung, und allen voran @Wissing, und die @fdp demontieren in diesem Augenblick die wohl wichtigste Ressource in der deutschen Klimatransformation: Vertrauen. #Lanz

[Jan Clusmann @JClusmann]

This is highly concerning and concerns all of us.

[Luisa Neubauer @Luisamneubauer]

Und da oben beim gelben Strich hat man in Deutschland - einem der Top 10 klimaschädlichsten Länder der Welt - befunden, es sei ein guter Zeitpunkt, das #Klimaschutzgesetz zur Unkenntlichkeit zu demontieren.

[Jan Clusmann @JClusmann]

@ADFC_Dresden @stadt_dresden Leider ist dieser Radweg alles andere als sicher, da nicht baulich getrennt + zwischen 2 Fahrspuren. Nur bauliche Trennung schützt Radfahrende...

[Luisa Neubauer @Luisamneubauer]

Ein historisches Urteil vom europäischen Gerichtshof! Wenn Regierungen Klimaschutz verschleppen, verletzen sie Menschenrechte.
Eine Zäsur für 1 europäischen Kontinent voller Staaten, die sich nicht an Klimaziele halten. Die Türen für weitere #Klimaklagen sind sperrangelweit auf.