Conversation
Worm Queen
desea@snug.moe@snacks it probably has enough grammatical errors in it’s training data to have “wrong” usages in there and show up tbf, especially for common errors so not rlly that shocking no?
1
0
1
@desea it's only purpose is to produce coreect or at least plausible sentences and the companies behind them put a lot of effort into the training data. It strikes me as pretty odd
2
0
0
Worm Queen
desea@snug.moe@snacks i mean, is a mixed case sentence implausible? it seems v plausible to me unless you have an underlying assumption of the llm adhering language rules not even native speakers adhere to 100%
1
0
1
Worm Queen
desea@snug.moe@snacks what constitutes “better” in your mind here? more adherent to an arbitrary set of rules? and why is that better inherently
1
0
0
0
0
1
Worm Queen
desea@snug.moe@snacks well it doesn’t “know” what correct is or isnt, it can only infer from training data and the inference just said yea thats most likely there
0
0
1