I experience the same, forcing me to edit the text first in a separate editor and then paste it into ChatGPT.
Following the bug report form:
Bug Report:
Steps to reproduce:
-
Preparation: Any recent version of mac OS with Safari. Activate the standard Japanese input method from the Keyboard section in the Settings app.
-
A sample text that you will try tying is: "京都の歴史を教えてください (please tell me the history of Kyoto, Japan)"
-
With the Japanese input method, type "kyoto", which is a romanised phonetic form for the first word "京都". You will observe that Latin characters you typed are converted automatically to Hiragana, Japanese phonetic characters, and then as soon as you completed typing "kyoto", it will be in the final form "京都". Observe the underline, which indicates that the part of the text is still in the Japanese editing mode.
-
Type a return key to confirm the result. (if you did not get "京都" and see different conversion such as "今日と", type space keys a few times to get "京都" and then type a return key to confirm the result).
Expected result:
The first word you typed, "京都", has now been confirmed, which is indicated by underline being removed. You can now continue typing the rest of the text in the similar way.
Actual result:
Often, but not always, ChatGPT sends the return key to the server and it results in sending the incomplete text, only the first word of the text you intended to type, to the server. The server starts to generate the response to the incomplete text.
Additional information
Browser: Safari 16.x
OS: mac OS 13.x
The symptom is not observed on other web pages with text input field such as Google, Bing, twitter, etc. It was once observed on Facebook and many people with similar environment suffered, and it was quickly fixed.