I didn't conduct any tests. I've been using gpt-4-1106-preview since its release, and it felt like a real deal. However, for the past two days, it seems like a fine-tuned 13B LLaMA2? It repeats itself and avoids answering the question you asked, going all around it instead. It definitely doesn't feel like the same quality anymore.
ChatGPT can now switch between different models on its own (dall-e, data analysis, ...).
For me it felt like it can also switch to 3.5 (to save on resources) and it sometimes misjudges the questions.
"Are you really sure?" seemed to have triggered a ChatGPT4 response, but seeded with that wrong knowledge of ChatGPT3.5.
The GPTs also don't seem to run on 4.0, but on 3.5 with greater context or something in-between?
ChatGPT can now switch between different models on its own (dall-e, data analysis, ...).
For me it felt like it can also switch to 3.5 (to save on resources) and it sometimes misjudges the questions. "Are you really sure?" seemed to have triggered a ChatGPT4 response, but seeded with that wrong knowledge of ChatGPT3.5.
The GPTs also don't seem to run on 4.0, but on 3.5 with greater context or something in-between?