Hard to overstate how much I'm enjoying having access to a weird collaborator that basically never says no to any of my stupid ideas
A super posh pelican with a monocle watching the Monaco F1
More like the first one please
Add a walrus
(The main difference between generative AI and a human illustrator is that I wouldn't ever dream of wasting a human's time with this stuff)
I used to have a strong policy of never saying "please" or "thank you" to an LLM because it felt like anthropomorphism, and also felt absurd
I'm experimenting with "please" now on the basis that the training data is full of examples of people getting more constructive replies through not being rude!
@simon sort of interesting that the AI is now teaching you how to talk to it.
IMHO, it is not completely unrelated from how we interact with inanimate objects too. We try to be "emphatic" with the tool, in the sense that we try to manage it in the way it should be managed. We know when we are "gentle" with it, or when we are using it outside its limits, and stressing it.
When I speak to ChatGPT, I think to some tensors, and I imagine that if I'm kind and I give positive feedback when it is on the right path, then there are less doubts in it, and it can explore further in the right direction, because it knows that it is the correct path. If it is not understanding what I mean, the discussion can become more "rude" and stressful, but in this case it can be a positive "stress", because it helps to be on the same page. Like with normal people.
So there is a very high correlation between how we should talk to a real person and to ChatGPT in an effective way. We are anthropomorphism LLM, but often because this is the correct way to use this tool, up to date.
Said this, there are for sure people that interact with ChatGPT, believing it is a real self conscious entity, and not only the best way to use the tool.
