Gpt-4o-search-preview does not support attachments?

API
1

there is some kind of contradiction, and I get an error when uploading an image to base64

Screenshot (1)
Screenshot (1)2540Ă—1256 333 KB

The image shows an interface of a prompt playground with functions to upload files or images, but displays a message indicating that the current model does not support attachments. (Captioned by AI)
The image shows an interface of a prompt playground with functions to upload files or images, but displays a message indicating that the current model does not support attachments. (Captioned by AI)2540Ă—1174 195 KB

2

does anyone else have this problem?

3

Doesn’t even have images as input, let alone new file attachments that on chat completions only work for PDF on the file search model.

Your choice of error:

  • Yes, base64 with data URL into Responses
    Invalid image URL: 'messages[0].content[0].image_url.url'. Expected a base64-encoded data URL with an image MIME type (e.g. 'data:image/png;base64,aW1nIGJ5dGVzIGhlcmU='), but got a value without the 'data:' prefix.

  • base64 into Chat Completions
    "Invalid content type. image_url is only supported by certain models."

  • try reducing the nesting, like Responses, or try internet url:
    Invalid type for 'messages[0].content[0].image_url': expected an object, but got a string instead.

So there are cases like going from image to internet search would take a pairing of two models, tools or automatic injection of the missing resource.

5

Let me retort:

If you’re getting irrelevant answers from an AI that has no understanding of the issue, no training on the API being discussed, nor a comprehensive context at all of what is being discussed when you are using the AI to make automated answers into a forum meant for real human interactions, it could be due to several reasons. Here are some common issues and how to troubleshoot them:

  1. Insufficient Context Provided to the AI:

    • Issue: The AI model requires adequate context to generate relevant answers. Without sufficient contextual information, it defaults to generic or unrelated responses.
    • Troubleshooting: Ensure that your implementation explicitly passes detailed context about the topic, previous interactions, user queries, or relevant forum threads to the AI. Adjust your prompt design to include comprehensive context.

  2. Lack of Domain-Specific Training or Fine-Tuning:

    • Issue: General-purpose AI models may lack specific knowledge or expertise about your forum’s domain or the particular API you’re referencing.
    • Troubleshooting: Fine-tune the AI model with domain-specific training data, documentation, or example interactions. Utilize specialized datasets or documentation relevant to your forum’s topic and the API in question.

  3. Inappropriate Prompt Engineering:

    • Issue: The prompt or instruction provided to the AI might be ambiguous, overly broad, or unclear, causing the AI to misunderstand the intended scope of responses.
    • Troubleshooting: Clarify and refine your prompts. Be explicit and concise about the expected output, context, and constraints. Experiment with structured prompts that guide the AI to provide relevant, precise answers.

  4. API Integration Errors or Misconfigurations:

    • Issue: Incorrect API parameters, improper integration, or outdated endpoints might lead to AI responses that seem irrelevant or disconnected from your intended use case.
    • Troubleshooting: Verify your API integration thoroughly. Ensure correct endpoint usage, accurate parameter passing, proper API keys, and compliance with the latest documentation provided by the AI provider.

  5. Outdated or Incomplete Data Sources:

    • Issue: If the AI relies on external data or knowledge bases that are outdated, incomplete, or inaccurate, its responses will reflect these deficiencies.
    • Troubleshooting: Regularly update and maintain the knowledge base or external resources the AI references. Perform routine audits of the AI’s data sources, documentation, and training materials to ensure accuracy and completeness.

By systematically addressing these common issues, you can significantly improve the relevance and usefulness of AI-generated automated responses in your forum, enhancing the quality of interactions and user satisfaction.

The actual answer is to keep unlabeled bot text passed off as a human response off the forum.

2 Likes
6

I use chat completions. Everything is working correctly with gpt-4o. But when I set gpt-4o-search-preview without parameters instead of gpt-4o, it doesn’t work. Then why do they write that images are possible in the input in this model? Also, it’s not available in playground.