Raw LLM Responses

I remember reading that OpenAI dumbed down GPT's capacity to sort through pdfs/docs in order to curb students using it for homework. They programmed it to make it much less helpful. Been using it since it first became public, and I noticed the transition where it started to suck on purpose. I've had so many such frustrating conversations where it completely ignores my requests, starts hallucinating bs or, and just straight up telling me to read the pdf myself. A solution I found is prompting "store text (content)", in a series using the content on the pdf/doc. It has a short memory and you are limited in the number of words, but it actually follows your prompts then..