Actions unable to parse Unicode escape sequences correctly

I have a API response the following JSON:

    "value": [
            "id": null,
            "changeKey": null,
            "birthday": null,
            "displayName": "\u532F\u8C50\u9999\u6E2F\u5BA2\u6236\u7D93\u7406",
            "emailAddresses": null

Which the displayName property should be 匯豐香港客戶經理. But the Actions will read this string into 盈豐香港客戶經理 or 豐富香港客戶經理 or 政賀香港客戶經理 or 測試香港客戶經理, … There are unpredictable.

This is not only reading the API response. When I send request to a Action with a API call, if the parameters contains Chinese or any non-ASCII characters, Actions will encode these words using Unicode escape sequences and send them to the API. Some words (not all) will produce wrong Unicode escape sequences as well.

I’m not sure how OpenAI deal with these Unicode escape sequences. If OpenAI use LLM to deal with that, that could be a serious problem because the result will not be predictable. That is not suitable to call functions. If OpenAI use some package to handle that, I think that package might be buggy. It should be replaced.


I believe the same issue has been recorded in this thread: Gpt-4-1106-preview messes up function call parameters encoding