Is this (still?) true? :eyes:
# compose
t
Is this (still?) true? đź‘€
l
ChatGPT uses an archive from 2021 for its training, and currently does not have access to the up-to-date web. I would be wary of anything it says about Compose.
For this specific method, it doesn’t autocomplete in Android Studio, and I don’t see it in the current source for SnapshotState.kt (where it used to be according to a Google search).
a
Also, it does make up BS sometimes. Like if I ask how to do a specific thing in Jetpack Compose (like custom height bottom sheet), it'll mention function arguments that don't actually exist and probably haven't existed, like
dragFraction
.
l
ChatGPT’s biggest challenge is that it’s always confident. I find that most output looks right at first glance, but it typically has some small error for anything more complex than a short function.
k
We need something to mark questions on these “answers” from bots
l
Supposedly, it’s possible to copy/paste the new API docs into a response, and it’s supposed to update its knowledge (for just that conversation, of course), but I haven’t seen this work well. It forgets after a few replies.
k
We need a new term for this overconfidence. Chatsplaining? Botsplaining?