mlazowik
They are not just an LLM answer, they are an (often cached) LLM summary of web results.
This is why they were often skewed by nonsensical Reddit responses [0].
Depending on the type of input it can lean more toward web summary or LLM answer.
So I imagine that it can just grab the description of the „car wash” test from web results and then get it right because of that.
https://text.npr.org/897692090
> Plastic also degrades each time it is reused, meaning it can't be reused more than once or twice.
There are even ISPs that do that: https://security.stackexchange.com/questions/157828/my-isp-b...
https://thenextweb.com/insights/2017/12/11/comcast-continues...
But definitely agree that cleaner code > performance in most cases. First make sure you really do need to squeeze that performance out.
They might even not publish the source, and if they do, what are the chances that anyone has audited it?