I asked three different commercially available LLMs the same question: Which TLDs have the same name as valid HTML5 elements? This is a pretty simple question to answer. Take two lists and compare them. I know this question is possible to answer because I went through the lists two years ago. Answering the question was a little tedious and subject to my tired human eyes making no mistakes. So…
not to defend AI, but that is an expected outcome if you think about how LLMs work.
It tokenizes language to run math on it and send back processed natural-sounding language
of course it struggled with “how many rs in strawberry”
because
would all present as different tokens of different length with different information spaces relative to them.