Chat completion message object - “name” issues

API Bugs
1

Consider the following message object:

{“role”: “user”, “content”: “My name is Bob”, “name”: “Bob”}

The “name” entry is lesser documented but can be used to keep track of different user identities in a conversation.

First issue: is this being deprecated or something? There used to be documentation on it but now all I can find is this:

That link is the documentation for “user” which is a parameter of client.completions.create. But what I’m talking about is “name” which is a key/value entry in the message object.

Second issue: The gpt-4-vision-preview model cannot access the information within “name”. In other words, when I put some text in the “name” entry and tell gpt-4-vision-preview to recite it, it can’t. All other models I’ve tested can do this (gpt-3.5-turbo-16k, gpt-4, gpt-4-1106-preview).

2

Thanks for reporting! We’re fixing the documentation issue; name is still supported on our models. However, gpt-4-vision-preview does not support it yet; we are working on fixing that too.

3 Likes
3

Thanks! Can you explain the difference in purpose between “name” and “user” (the link I included above)?

4

It appears the “user” field is not a per message field, but a per request field. It seems to primarily be used for safety tracking. Ie, If there is a violation of OpenAI’s policies, they may be nice and not ban you entirely if you have provided user ids, and they could ask you to ban that user from your service instead.