new release (codex on cloud) has significant regression with context retention and code generation
What version of Codex is running?
cloud version
What subscription do you have?
pro
Which model were you using?
cloud version
What platform is your computer?
cloud version
What issue are you seeing?
the new release on cloud is significantly regressed. When a promot is given it does not follow the instruction, when asked or provide additional context in a follow up promot those prompt instructions and responses disappear and then the changes go back to the original question that was asked. It's probably the worst performing code generation tool at this point as it does not retain context nor does it adhere to teh task being asked.
What steps can reproduce the bug?
give it a prompt to ask questions like "what optimisation can the code do for x,y,z"...it will provide some responses...ask it clarifying questions to determine which ones are actually valid...those responses disappear once u ask a follow up question...the follow up question then reverts back to explaining the base question..
What is the expected behavior?
follow insutrctuins and maintain context. basic 1:1, was not expecting such a broken release
Additional information
_No response_
This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗