The more extensively you use the AI, alongside offering it the provided information you give it the more it seems to lose track and focus on the actual content provided and instead fills it up with its own content and idea as to what it sees fit based off of the discussion at hand. I offered it information to analyze and utilize for my purpose and after awhile it just made up its own content.
#ChatGPT forgets?
wraith ember
Bug Report:
Steps to reproduce:
- Give it documents to write a DBQ
- Give it specific requirements to actually meet the standards of a proper DBQ such as contextualization, telling it to meet the rubric and prompt's standard, not mentioning the document but rather indirectly, etc
Expected result:
Supposed to follow through
Actual result:
Had ended up forgetting the entire prompt and documents I had provided and seemingly made its own up in thin air and can't seem to recall what the original prompt/documents were.
Additional information
Browser: Chrome
OS: Windows 11
brazen hull
AFAIK, in long conversations the AI will only remember a certain amount of text. There is still a maximum amount of tokens accepted for the input. I tried to ask it how much, but I never got a good answer.
wooden pivot
Have replicated on GPT-4, May 24th edition.
Expected result:
GPT should recall two prior conversation posts prior. Less than 100 lines of text between.
Actual result:
GPT appears to only parse the previous conversation post, and is ignorant of prior conversation.
Additional information
Browser: Firefox, Chrome, Edge, Brave, Opera.
OS: Windows 10, Android.
Will be reporting affected conversations shortly.
Workaround
In lieu of new posts in the line of conversation, I have found it is possible to work around this issue by editing the topmost post with new information. Tedious, but functional.