Could this be simply explained by retrieving from the original that’s present in the context? You’d want to open a separate ‘session’ in ChatGPT for this, and see if the ‘recovery’ is still lossless.
1 Like