I upload a pdf with all my artworks (16 and ask teh AI to either review or make comments about every single one, it does it for 4 or 5 but then starts making up stuff and artworks that aren't the ones in the pdf
#AI stops reading document halfway through
midnight crypt
spare nacelleBOT
Hi @midnight crypt. I'm kapa.ai, your friendly support bot. I will try to answer your question by searching through the knowledge base. :wave:
It seems like you're encountering an issue where the AI stops processing your document halfway through. This could be due to a few reasons:
-
Token Limit: The AI models have a maximum token limit. For instance, the GPT-3 model has a hard output limit of 4k response tokens. If your document exceeds this limit, the AI might not be able to process the entire document. You might need to break down your document into smaller parts and process them separately.
-
Ambiguous or Unclear Images: If your PDF contains images of your artworks, the AI's ability to interpret and comment on them might be affected if the images are ambiguous or unclear. The model will do its best to interpret it, but the results may be less accurate.
-
AI Interpretation: The AI might interpret the "stop" token in your document as a signal to end the output. This usually happens when the AI thinks your document is complete. You might need to adjust your prompts to make it clear that the document is not complete and that the AI should continue processing it.
You can try using OpenAI's playground, namely “assistants” where you can pre-prompt an agent and attach the pdf file to “vector file storage”. Then in your prompt say something like "review the attached PDF file and comment on each artwork".
Remember, it's important to set clear and specific prompts for the AI to ensure it understands your task correctly.
Sources:
- Forum Post: Max_tokens not set, truncated return with "finish_reason": "stop" (https://community.openai.com/t/725740)
- Forum Post: Force ChatGPT to read an entire long PDF (https://community.openai.com/t/788955)
- Help Article: How do the image capabilities handle ambiguous or unclear images? (https://help.openai.com/en/articles/8400551-image-inputs-for-chatgpt-faq#h_d8bcc787dd)