ChatGPT not able to analyze text anymore?
Initially tried importing .pdf and .doc files, error. Tried pasting them as .pngs and .jpegs… still errors.
This is with a ChatGPT team subscription btw.
If you are sending an image, you may specifically need to state “use your built-in gpt-4 machine vision to extract and describe the text contents of my attached picture”.
The larger problem is likely the AI emitting python functions when it has no use, even writing code that serves no purpose. It seems OpenAI put tool-calling on a hair trigger, and then the AI must make up something to write to the following function parameters.
Putting the PDF into knowledge retrieval should show “reading documents”. Then the AI can search besides what is loaded into context.
Here’s a prompting technique to get AI to talk about the backend PDF parsing that is loaded to AI memory, and also perform a searchable text extraction and return it.
techies: this takes four individual assistant agent outputs to fulfill
to=myfiles_browser code=open_url("file-gTcZntgwrNCzsEMyc3gmrGt9")
The document presents a novel approach to enhancing few-shot learning called FewXAT, which focuses on explainable hard attention finding. This method uses deep reinforcement learning to identify crucial data segments within inputs, improving model performance by concentrating on the most informative parts. The research demonstrates FewXAT's effectiveness across various benchmark datasets, showcasing its potential in improving computational efficiency and interpretability in machine learning models, especially in scenarios with limited training data.
I will now extract the text from the first page and prepare it for download as a .txt file.
to=python code=
from PyPDF2 import PdfReader
# Open the PDF file
reader = PdfReader("/mnt/data/attention-paper.pdf")
# Extract text from the first page
first_page_text = reader.pages[0].extract_text()
# Save the extracted text to a new .txt file
with open("/mnt/data/first_page_text.txt", "w") as text_file:
text_file.write(first_page_text)
# Provide the path to the saved .txt file for download
"/mnt/data/first_page_text.txt"
The text from the first page has been successfully extracted and saved to a .txt file. You can download it using the link below:
[Download the extracted text](sandbox:/mnt/data/first_page_text.txt)
1 Like
Wow, super helpful! Thank you so much for sharing this solution! I had this same issue. How did you know this was the fix? I have difficulty getting ChatGPT to do anything for me, but this solution successfully got GPT to extract information from my PDF for the first time ever. Getting GPT to review anything has been a pain, are there similar solutions for .docx/.xlsx/.pptx/etc.?
This is happening to me too. It also will not create docs or tables anymore. This so annoying. It’s like it’s getting dumber!