Hi everyone! I’m using ChatGPT to classify thousands of scientific papers based on their abstracts and titles. I’ve discovered that ScholarGPT (which I think is based on normal GPT-4), is a perfect fit for my needs. It’s more effective than ChatGPT-4o, though that one is still pretty good—just about 10-20% less accurate in my experience. However, I’ve noticed that both models seem to struggle more with long chats. For context, I use two types of prompts: the first one contains the instructions (450-500 words, 3500-4000 characters), and the other prompts include groups of 10 abstracts (2500-4000 words, 15000-25000 characters).
I could really use some help understanding the token limits for ScholarGPT and ChatGPT-4o:
- What’s the maximum number of tokens (estimated in words) per user message?
- What’s the maximum number of response tokens (estimated in words) without causing issues?
- What’s the recommended number of tokens (estimated in words) per chat window?
- Which model does ScholarGPT use: ChatGPT 3.5, ChatGPT-4, or ChatGPT-4o?
I prefer working through the browser and have no interest in APIs or console work. Any advice would be greatly appreciated. Thanks a bunch!