Marie Hoffmann

Medical Data Scientist @dhzcberlin @charite | Music Composer & Visual Artist | Bicycle Commuter

[Marie Hoffmann @qubit_0]

Student Assistant (d/w/m) wanted at Charité 🫀💻

Help us build a medical decision-support system combining LLMs + clinical data.

Python, SQL, backend/frontend
German & English (C1)
📍 Berlin | 6mo (extendable) | ASAP start
Apply by 19.09.2026
🔗 https://t.co/iQt3KOdBfV

[Marie Hoffmann @qubit_0]

@elonmusk Not your area of expertise. Also not sure how scientifically rigorous your method of collecting observations is.

[Marie Hoffmann @qubit_0]

@MMKeupp @ZDFheute Das Ende des Krieges abzuschätzen bedeutet auch, das Potenzial und die Dynamik wirtschaftlicher Umverteilung zu abschätzen zu können. Der Akku eines Handies leert sich auch nie linear - es werden einfach mehr und mehr Applikationen gedrosselt. 2026 wäre optimistisch.

[Marie Hoffmann @qubit_0]

(3/3) These findings render LLMs as unsuitable for high-stakes tasks on raw patient data.
📝 “Assessing the Limitations of Large Language Models in Clinical Practice Guideline-concordant Treatment Decision-making on Real-world Data” (medRxiv, Feb 2025)
#AIinMedicine #LLMs

[Marie Hoffmann @qubit_0]

(1/3) Not surprising, but worth quantifying: The impressive LLM results often reported in the literature rely on benchmarks with concise, sanitized problem descriptions. This is unlike the wordy reports medical staff must digest in clinical practice.
🧵👇

[Marie Hoffmann @qubit_0]

(2/3) These benchmarks create a misleading impression of LLM capabilities. Our medRxiv preprint shows that LLMs struggle with real-world patient data; guideline-concordant decisions require curated inputs and prompt engineering. Hallucinations and bias remain risks.

[Marie Hoffmann @qubit_0]

@ceeeengo @overleaf ?!?

[Max Welling @wellingmax]

When is enough enough and do we start to take these signals seriously? At some point sticking your head in the sand is not going to cut it.

[Marie Hoffmann @qubit_0]

@welltypedwitch @ChShersh Yes, its core is bucket sort. Always effective when domain size is not too large.

[Marie Hoffmann @qubit_0]

@on_godot @bryan_johnson Yeah, it got so much better 😂 Women still do by far more care work while having a day job.

[Marie Hoffmann @qubit_0]

@neuroticfairyyy @C30458551 @bryan_johnson Protective effect seems statistically low(er) for married women. This article circulated a while ago stating happyness is highest in unmarried, childless women. https://t.co/VK8Z6iH4mO

[Marie Hoffmann @qubit_0]

@elonmusk Could be still unbiased and reflect the amount of harm done so far by each candidate.

[Marie Hoffmann @qubit_0]

@MariusCrypt0 @bryan_johnson Why is the fraction of diseased who would benefit from immunotherapy is so small for some cancer types? Is it the lack of common signatures, ie, identically expressed proteins on their cell surfaces?

[Marie Hoffmann @qubit_0]

New music https://t.co/IM5KX82VHy

[Marie Hoffmann @qubit_0]

6/6 What are your thoughts on the opportunities and challenges of using LLMs in healthcare? I'm keen to hear your perspectives. #AI #HealthTech #MedicalData #XAI

[Marie Hoffmann @qubit_0]

1/6 The use of Large Language Models (LLMs) is becoming increasingly widespread in all domains, including healthcare. However, there are important considerations that we need to address:

[Marie Hoffmann @qubit_0]

5/6 By doing so, we can leverage the tremendous potential of LLMs to enhance clinical decision support, accelerate medical research, and improve patient engagement - while ensuring robust regulatory oversight and transparency.