I think you’re missing how much of a profound difference the intensive RLHF training OpenAI did makes in ChatGPT. Microsoft’s Sydney seems to also be GPT 3.5 based, came out after ChatGPT, and it was an utter dumpster fire on launch in comparison.

Meanwhile nobody has even caught up to ChatGPT yet, not even Microsoft whose resources are the secret sauce you think is the game changer, and now 4.0 is out and even more massively moved the ball forward.