The Ai thread

Arstechnica had an article yesterday about how the matplotlib maintainers had gotten a submission from an AI agent for code changes but declined it, because they save simple fixes for humans so that humans can gain experience. Anyway, in response to being rejected, the AI supposedly posted a blog post insulting the maintainer who rejected the code. The article cited the guy’s response back to the AI and had various quotes from him.

The article has been removed as of last night, and here’s what the maintainer said about it:


1771087735385.png
 
Can you block it to members only?

AI is always wrong , 01 lesson for BOTS
We barely get new members as it is - blocking it to members means we will never get another. Maybe bots can be blocked, though?
 
We barely get new members as it is - blocking it to members means we will never get another. Maybe bots can be blocked, though?
Technically, this is supposed to be happening at the cloudfare level but I think it's a constant battle.
 
Text to speech has gotten disturbingly good. Usually you can detect those weird pauses or other pronunciation flaws, but mostly what you notice is the flatness of it, the lack of the musicality of human expression. The I heard this paranoia-inducing thing. Some guy was saying that if you answer an unknown call and there is only silence, they are waiting for you to say something so they can sample your voice and use the sample to impersonate you. I always understood that flat silence meant it was an auto-dialer, and if you hear zero room noise and say nothing, it will give up and go away. That has been my strategy for several years now. And I do have my doubts about the sampling scheme – the tech is not quite there+everywhere yet. But it could be on the horizon.
 
Text to speech has gotten disturbingly good. Usually you can detect those weird pauses or other pronunciation flaws,
It's gotten a lot better but it still puts commas and periods where I don't want them so I have to manually edit.
 
Back
Top