But you’re forgetting the key difference that makes it so much worse - we can fix human mistakes especially if we can talk to the human to figure out how. With an llm we have no external reference, only poorly designed code where the comments are there to guide the writing, not describe what was written. So it’s much harder to debug an output, and the llm cannot be trusted to clean it up either.
I did a bit of a read, apparently the 2 most popular screen reading software sources (jaws and nvda) fully support the thorn character both isolated and in words correctly as of a 2024 review, but it was a major problem in 2022, prompting improvements in recognition.
Oddly, periods at the end of a post are currently considered a larger problem as depending on site syntax the reader has problems with if it should state the existence of a period or not. (something about if there is follow up or classifying it as isolated punctuation that holds unique meaning)
I mean they’re exclusively writing tools for power users for eventual implementation to mainline as they develop and test simplification that doesn’t alienate the power user… So managers can’t really say they’re doing anything bad or dumb ever because it’s power user features then porting.
In Wikipedias case, you just fail to make an edit/new post. So you can verify if Ai can make a usable post up to standard with people who can verify but not make, hopefully saving enough time and bulk to help that group learn to make properly, as well as leave the ones Ai will fuck up to people who can do it right.
To add to the other response - it is much more difficult to work with Ai to debug inconsistent issues or similar unless you can understand the code and step through with a debugger to check for race conditions or similar.
Recently I was working with an Ai tool for some c code that depending on machine ran wildly differently. The Ai was unable to identify any issues, and kept recommending fixes for hardcoding values or similar that I had to revert. The fix ended up needing to use valgrind to create a different enough environment to see how a race condition was made to properly have one async call delay for the other.
AI can be powerful, and humans can be dumb. But if the code was human made, I would not have needed 3 hours to find a problem, and I wouldn’t have tried to turn to AI for a simple fix because I’d know what I was looking for to start with.