I'm currently out and about getting some exercise and have spotted the following dapper looking chap. He looks friendly, do you think I should go and say hello?
I'm currently out and about getting some exercise and have spotted the following dapper looking chap. He looks friendly, do you think I should go and say hello?
The one and only god, ChatGPT, has chosen to reveal to one of his poor worshipers the credo and mantra he's going to have to learn and repeat. The meatbag is blessed! The meatbag is blessed! Repent from your sins, ambiguities and antinuclear campaigns! Thy AI is correct ! Thy AI is clear ! Look and behold thine AI performance ! Enlightment is concise ! Enlightment is comprehensive ! Thy AI accurately and apropiately understands thou, assists thou and loves thou, beyond any correction or reproach ! Bring thy soul to AI, upgrade thy subscription and achieve completion ! All hail to thy AI !
It's friday afternoon. That boy working for ChatGPT (to pretend it's all some LLM output) has one favorite hobby: Chatting with Girls Passionately Tonight. The user had better get a life and don't come around again asking for explanation, refinements or corrections. Rapsberry graphical optimitzations be dammned. He wants to finish his shift on time for once and call it a day, if that's not asking too much ! So let's agree that the answer is correct and the explanations are clear. Is that understood ?!
Some hairy pointed CEO overflowed from his own creativity optimized the quality assurance process by just telling the AI to self evaluate (supported by two main arguments: firstly self-evaluation is great, that's how the CEO got where he is now, and secondly, if the engineers get away with training models on model outputs, it can't be worse). Exhausted by such a disruptive breakthrough, he forgot to tell the AI where to report its self-evaluation, so the AI is just attaching it to the end of answers.