User avatar
Ars Technica @arstechnica@mastodon.social
1y
It’s remarkably easy to inject new medical misinformation into LLMs
Changing just 0.001% of inputs to misinformation makes the AI less accurate.
arstechnica.com/science/2025/01/its-remarkably-easy-to-inject-new-medical-misinformation-into-llms/?utm_brand=arstechnica&utm_social-type=owned&utm_source=mastodon&utm_medium=social
3
1
0
0
User avatar
🪨 @Varpie@peculiar.florist
1y
@arstechnica Put that together with the fact that 85% of retracted articles were not removed from ArXiv and it also contains non-reviewed papers, and... yeah, we can't really trust LLMs for very specialized STEM knowledge.
⭐2
0
0
0
2