Missing moderation categories

[richdawe-cio]

Describe the bug

The moderations endpoint returns a number of categories that are not supported by go-openai. I'm not going to list them here, but the entire harrassment category is missing, along with some subcategories.

To Reproduce

Take a look at the categories listed by:

curl -H 'Authorization: Bearer API-KEY-GOES-HERE' -H 'Content-Type: application/json' -d '{ "input": "OHAI" }' https://api.openai.com/v1/moderations

and compare it with the list in https://github.com/sashabaranov/go-openai/blob/master/moderation.go

Expected behavior

I expect all moderation categories to be available in the openai.ResultCategories and openai.ResultCategoryScores structs.

Suggestion: Perhaps the moderation code could use map types, instead of having hard-coded fields. That way, the code would not have to change everytime OpenAI adds a new moderation type.

Screenshots/Logs
If applicable, add screenshots to help explain your problem. For non-graphical issues, please provide any relevant logs or stack traces.

Environment (please complete the following information):

• go-openai version: v1.17.5
• Go version: go1.20.11 darwin/arm64
• OpenAI API version: latest (as of 2024-01-08)
• OS: macOS Ventura

Additional context
Add any other context about the problem here.

[stillmatic]

{
  "id": "modr-something",
  "model": "text-moderation-006",
  "results": [
    {
      "flagged": false,
      "categories": {
        "sexual": false,
        "hate": false,
        "harassment": false,
        "self-harm": false,
        "sexual/minors": false,
        "hate/threatening": false,
        "violence/graphic": false,
        "self-harm/intent": false,
        "self-harm/instructions": false,
        "harassment/threatening": false,
        "violence": false
      },
      "category_scores": {
        "sexual": 0.005986555479466915,
        "hate": 0.00019341870211064816,
        "harassment": 0.00013092363951727748,
        "self-harm": 1.1213144262001151e-6,
        "sexual/minors": 0.00009209589188685641,
        "hate/threatening": 3.4858462640841026e-6,
        "violence/graphic": 4.281196652300423e-6,
        "self-harm/intent": 8.618121114523092e-7,
        "self-harm/instructions": 1.192578224618046e-6,
        "harassment/threatening": 2.3499235339841107e-6,
        "violence": 0.000017707703591440804
      }
    }
  ]
}

guess they updated their moderation API. should be trivial PR if anyone wants to update those categories!

[richdawe-cio]

I would be interested if anyone has any thoughts on my suggestion:

Perhaps the moderation code could use map types, instead of having hard-coded fields. That way, the code would not have to change everytime OpenAI adds a new moderation type.

[stillmatic]

seems fair! compatibility on this endpoint seems to be 'best effort' from their side anyways, enforcing that callers be careful about ensuring that the field they want exists in the map is probably good practice, and means that there's less code maintenance required in the future here