Other than by affecting the output length the complexity of the input won’t affect how likely it is to succeed, as each token always takes a fixed time to generate and there’s no way for the AI model itself to crash

At sufficient scale everything becomes a daily occurrence. Hardware failure e.g.

Yes, but it’s not related to the complexity of the input

Given deterministic everything sure, but nothing really is. Perhaps GPU 7 has a very toasty neighbor and is slower.

What I’m saying is it’s not correlated to the complexity of the input, because GPT models don’t have a built in way to say “let me stop and think about this”, it’s a fixed amount of computation per token

I first noticed it 30min ago when GPT 4 was down for me.