Generative AI Fully Explained by Daniel Wolf - Journal Club [13]

Описание: In this video Daniel Wolf an AI researcher at the University of Ulm Germany explains how generative AI multi-modal models like ChatGPT, Gemini work, their data privacy, hallucinations, and much more.

Join us: https://clinicalaiacademy.substack.com/

References:
Medical Knowledge: ChatGPT passes German State Examination in Medicine: https://doi.org/10.3238/arztebl.m2023... (Peer-reviewed Paper)

Medical Knowledge: ChatGPT outperforms MDs on medical Reddit questions and is more empathetic: https://doi.org/10.1001/jamainternmed... (Peer-reviewed Paper)

Image-Text: MLLMs fail at Spatial Reasoning: https://openaccess.thecvf.com/content... (Peer-reviewed Paper)

Image-Text: MLLMs Prioritize Memorized Knowledge over Image Input: https://doi.org/10.1007/978-3-032-049... (Peer-reviewed Paper) and https://arxiv.org/abs/2512.04238 (Preprint Paper)

Bias: AI for Recruiting based on CVs: https://www.reuters.com/article/world... (News Article)

Bias: More mistakes on female images in X-ray classification tasks: https://www.science.org/doi/10.1126/s... (Peer-reviewed Paper)

Bias: Intentional Contamination with Russian Propaganda: https://www.newsguardrealitycheck.com... (News Article)

Bias: DeepSeek performs worse on coding tasks when China critical is added to the prompt:    • CrowdStrike: A Deep Dive into DeepSeek and...   (News Video)

MLLMs are overconfident: https://arxiv.org/abs/2511.04869 (Preprint Paper)(Edited)

Video for educational purposes only.

Timestamps:
00:00:00 Intro
00:00:32 Presentation start
00:38:00 Discussion
#ai #medicine