AI Mythbusting

 

EP. 09

In this episode, we tackle some common myths about how generative AI works, why this is the case, implications for healthcare and some quick fixes. These myths include 1) that LLMs can explain their reasoning 2) that LLMs can express uncertainty, 3) that LLMs can a) do maths, b) manage temporal data c) apply guidelines d) handle negation and finally that 4) that AI will replace clinicians.

02:00 Technical update - DeepSeek, other new models

10:00 - AI Mybusting

  • 15:50 - LLMs can explain their reasoning

  • 21:50 - LLMs can express uncertainty

  • 26:40 - LLM blindspots

  • 41:50 - AI will replace clinicians

Some resources and papers we discuss: 

McCoy LG, Swamy, R, Sagar, N. et al :Do Language Models Think Like Doctors?” medRxiv https://doi.org/10.1101/2025.02.11.25321822

Cabral S, Restrepo D, Kanjee Z, et al. Clinical Reasoning of a Generative Artificial Intelligence Model Compared With Physicians. JAMA Intern Med. 2024;184(5):581–583. doi:10.1001/jamainternmed.2024.0295

Griot, M., Hemptinne, C., Vanderdonckt, J. et al. Large Language Models lack essential metacognition for reliable medical reasoning. Nat Commun 16, 642 (2025). https://doi.org/10.1038/s41467-024-55628-6

Ahn, Janice et al. “Large Language Models for Mathematical Reasoning: Progresses and Challenges.” ArXiv abs/2402.00157 (2024) https://arxiv.org/abs/2402.00157

Fatemi, Bahare et al. “Test of Time: A Benchmark for Evaluating LLMs on Temporal Reasoning.” ArXiv abs/2406.09170 (2024): https://arxiv.org/abs/2406.09170

Wallat, Jonas et al. “A Study into Investigating Temporal Robustness of LLMs.” ArXiv abs/2503.17073 (2025): https://arxiv.org/abs/2503.17073

Zondag AGM, Rozestraten R, Grimmelikhuijsen SG, Jongsma KR, van Solinge WW, Bots ML, Vernooij RWM, Haitjema S. The Effect of Artificial Intelligence on Patient-Physician Trust: Cross-Sectional Vignette Study. J Med Internet Res. 2024 May 28;26:e50853. doi: 10.2196/50853. PMID: 38805702

https://medium.com/@avanib28264/no-elephants-in-the-room-ai-seems-to-think-otherwise-7d70d6a7d5a4

 
 
Previous
Previous

Introducing In-context

Next
Next

Algorithmic Bias