vaxvms Asking for a friend… Can AI perform QA on itself?
Yes, while I can’t speak to what Gemini is up to, AI performing QA on itself can be evaluated with ChatGPT especially when comparing the results of 4, 4o, and 5 but the question is whether or not the results are conveyed to the developers which are affecting these engines.
If you get an answer in ChatGPT, ask it to compare the results that would have been provided with all three models. You’ll find the answer is actually different for some questions. When you criticize how bad the answers are with GPT-5 vs. GPT-4 or GPT-4o, it will offer to customize the responses in 5 to better conform to the 4/4o model experience and if you’ve signed into your account, you’ll get slightly better responses with 5 (but still not as optimal as 4/4o).