RealFeedApp OLMo 2 does not know about 291 retracted papers
The OLMo 2 model failed to identify any of the 891 requested papers as retracted, even though 7.7% of the passages it read contained mentions of retractions.
- Field
- training language models
- What they did
- Researchers tested the open language model OLMo 2 on its ability to recognize the retraction of scientific papers using data from 291 retracted works.
- Why it matters
- The model showed that 7.7% of the text passages it read mentioned the retraction, but it failed to correctly identify the status of other papers because it did not know their authors or journals.
#llm#retracted papers#data quality#almo
