It's remarkably easy to inject new medical misinformation into LLMs

The paper: [Medical large language models are vulnerable to data-poisoning attacks](https://news.ycombinator.com/item?id=42640260)