The Estonian Language Institute (EKI) tested artificial intelligence models.
The models do not recognize Russian propaganda. Some models often repeat it.
When a user asks a loaded question, the models respond with propaganda. The tests were conducted in Estonian, English, and Russian.
Cheaper models make more mistakes. Better models make fewer mistakes.
In Russian, the models perform worse. They make up to 15% more mistakes.
An EKI scientist says the problem can be solved.
EKI also tested Estonian. The results were poor. Some models do not handle Estonian well.
On the EKI website, you can view the models' results. This helps in choosing better models.