Today, I was using Chat to help me review family member vocabulary with simple fill-in-the-blank drills. I instructed it to give me sentences describing a person whose family title I would need to correctly write in.
Example exercise: My mother and father are my ___. Example answer: Parents ✅
The exercises were straightforward, and went through brother, sister, aunt, uncle, etc. However, we got to this example:
My brother’s son is my ___.
I answered: Nephew.
Chat marked it as wrong and told me the answer was grandson. Obviously that was incorrect. It did the same thing on a similar question for niece. I asked Chat to check its work on those exercises, and only then did it correct itself.
I am not doing these exercises in English, but in a Slavic language. I will provide translations below just in case (TL) Slovene. But I was very shocked by how this model could get such a simple, logic-based question wrong.
Has anyone else noticed such simple errors in their studying? And if so, how often would you say they occur? Are there any exercises you have found that tend to produce more errors? And it makes me wonder if this is just AI models struggling with languages outside of English?
I think AI is a great tool especially for drilling vocabulary, but this has really tanked my confidence in using it. Yes I caught this error, but the only reason I caught it is because this was more of a review drill - I am just trying to get faster at recalling these terms rather than learning them for the first time. I shudder to think if I had no exposure prior to this session, perhaps I wouldn’t have known to double check and then I would have learned the incorrect vocab!
(Translation of the specific exercises in question:
Sin mojega brata je moj ___. “Vnuk.”
Hči moje sestre je moja ___. “Vnukinja.”
However the correct answer should have been “Nečak” and “Nečakinja” respectively.)
[link] [comments]
from Language Learning https://ift.tt/Emo15He
via Learn Online English Speaking
Comments
Post a Comment