It won't get noticeably better because the underlying idea is very limited
Just saw this and yeah, i believe this is true. AI writes shit code. And coding exercises/tests are all very formulaic, perfect for a context free network
bsky.app/profile/mattround.com/post/3lgxbbagw5k2r
(
simbosan is too tired, Thu 30 Jan 2025, 12:45,
Reply)
I reckon in 10 years time you'll wonder how you could have ever thought this
but I guess we'll both have to wait and see how it plays out.
I've been listening to interviews with recently employed devs who worked for chatgpt and other ai mobs talk about their experiences with it. They say they're losing oversight over how to comprehend how the models are advancing, as processing capacity and data is added it's developing capabilities in a whole raft of areas they don't have the metrics yet to measure. And a surprising number of people with direct involvement with the project have come to the view it should be stopped.
(
cumquat may will not be commenting further on the allegations, Fri 31 Jan 2025, 0:04,
Reply)