> Where are all these verbal tics coming from and why is it so hard to get rid of them?
It’s a side effect of post-training for effectiveness and efficiency at technical tasks.
Over time the models learn to pack as much information as possible into their available context window, because that’s one way to increase the effective intelligence.
Humans do this too with industry jargon, dense tech-talk, etc.
We have a limited capacity so packing it densely maximises what we can do with it.
If you’ve ever heard a “non technical” manager complain about the terminology in an IT meeting — this is why.
Yeah that was what I was most worried about when I read the top comment here. I found the use of language a feature not a bug. I don’t care how good it reads. If I can communicate with it concicely it’s enough to get my work done. I don’t hate the language for copy either, but yeah different users, different problems.
Makes sense. Then maybe we would need a separate simpler LLM trained on UI copy and good UX to decide this stuff and let frontier models do the implementation.
But who has both the compute power and the motivation to do such a thing?
I guess I'll just continue rewriting the UI one word at a time for the time being.
Each time you re-write keep a copy of the before and after with some notes on why. Then with a few good examples of this turn it into a skill to review/fix new UI copy.
Believe me I tried that. I tried giving concrete examples, I edit a 50 line changelog and tell it to learn the style and write another changelog for another app similar to it, I built a whole system on a Vale linter detecting these claudisms with hooks to rewrite them in my voice which I've trained from my thousands of hand written UI copy and blog writing.
Nothing works. This style of writing is deeply ingrained into these models.