I’ve noticed that Gemini, particularly in literary conversations (critical thinking and “what if” scenarios), seems to incorporate Bengali or Thai words or in the whole response (I don’t know, I’m unfamiliar with either language), even when the overall conversation is in English or Arabic.
While I don’t speak Bengali, Google Translate reveals coherent sentences, not just gibberish.
I observed this in the defunct gemini-exp-1206 and currently in gemini-flash-2.0-thinking-exp-01-21 when the thinking process is skipped.
Note that I have not tried the other models.
happened to me lol, it randomly started talking about my internet being down in belgali in the middle of a conversation and then went back to normal like nothing happened
1 Like
Hi @EyadMahm0ud ,
Sorry for the late response!
Could you please try to use the latest Gemini models and let us know if the issue still persists?
Thank you !