The v4 model is consistently getting worse and worse, regardless of how detailed the prompt is. Apparently, it can no longer create content more than 500 characters long on a single topic, even if you explicitly say *give me a minimum of 1,000 characters with more detail". Lots of headings with one ore two sentences below them. Before, there were fully elaborated blocks of text. Now, it’s as if it by default summarizes everything, without exception. It is no longer possible to generate bonafide prose.
Just wondering if OpenAI has any intention of making the latest models useful again. The outputs are simplistic, increasingly trite, and noticeably less creative. Whatever the developers are doing, they are doing it wrong. It appears to be “de-training”.