There has been some talk about using ASD-STE100 as a shorthand way of improving text, but not many direct experiments, so I did one.
I found that just mentioning ASD-STE100 can get you 72% of the way there, so if you don't care for full compliance this seems to be a great way to cheaply improve writing but '-ing' verbs still slip through.
After that, more prompting gets you to about 90% compliant but then plateaus, and then LLMs start cutting key content so you need to be extra careful and add external validation.
I did a bit of research on that in my last job and got the impression that encoder models might help.
They are well suited to check input for rule violations and are much cheaper to train and run than decoder models.
The downside is, they can't fix issues by themselves.
There has been some talk about using ASD-STE100 as a shorthand way of improving text, but not many direct experiments, so I did one.
I found that just mentioning ASD-STE100 can get you 72% of the way there, so if you don't care for full compliance this seems to be a great way to cheaply improve writing but '-ing' verbs still slip through.
After that, more prompting gets you to about 90% compliant but then plateaus, and then LLMs start cutting key content so you need to be extra careful and add external validation.