I talked for some time but chat gpt misspells words in urdu language and many sentences are also wrong according to rules. There have also been mistakes in Masculine and Feminine. Even after making corrections, the same mistakes were repeated, so there is a need to make a lot of improvement for conversation in Urdu. Sometimes the sentences get confused and the point is not understood. This about 3.5 version.
The temperature setting that ChatGPT uses internally may be too high for the perplexity of less common languages in the training data.
You could reinforce Urdu first by pasting some large texts in the tone of the language you want it to understand. Make a category header like “// language style and usage training //” in Urdu so AI knows it is not directly something to answer about.
The API, and OpenAI’s platform.openai.com playground may be a place for testing the models directly to see if lowering the temperature can improve generation where there are improper bits of language. You can then program an application to do specific tasks using that technique. API requires payment for data use if your account is older and your API free trial has expired.
It would be nice if the parameters were exposed in ChatGPT — but it would also take re-educating about 100 million users.