Hi, in my previous thread @OpenAI_Support asked for additional information but the thread is uneditable and unreplyable so I’m adding the information here.
Environment
- Surface: Android, Windows, Linux
- Browser or app version: Firefox, Chrome, Chromium, Opera
- Affected feature and model, if relevant: Chat box UX
Bug
GPT does not see markdown formatting but it is forced upon the user breaking simple sequences like equations and code blocks. Escape characters don’t work (and would confuse GPT anyway since it does not see the markdown, it is only a client side feature that cannot be turned off)
e.g. if you type into the chatbox
calculate 3*5*9 + 7*6 + 9*cos(2*3)*e^x where x = 5
it will show the user:
calculate 359 + 76 + 9cos(23)e^x where x=5
which is obviously a completely different question. though gpt will still interpret the original input correctly
Reproduction
Type this into the box (do not paste):
what is 3*5*4
Notice that the user sees “what is 354” and cannot fix this using conditional formatting even though GPT itself sees exactly what is written above
Type this into the box (do not paste. press shift+return on the keyboard instead of literally typing the characters [shift+return]):
```print("hello world!")``` [shift+return] this should no longer be in a code block
Notice that the user sees this:
print("hello world!")```
this should no longer be in a code block
even though GPT sees the contents of the first box, the user sees input that isn’t inside a code block inside one as well as visible backticks. As soon as one uses mathematical equations or code blocks, the user cannot see what input is actually being transmitted to GPT. There is no way to disable this functionality and it makes code and log analysis and arithmetic/probabilistic analysis difficult to use across all platforms. I am currently using my own greasemonkey script to override the text box because it is unusable.
Diagnostics
Exact error or screenshot:
When it happened - Every day. How often: every single chat.