My boss cannot understand that Claude is physically incapable of learning concepts.
Me: Claude made this mistake.
Boss: Okay, so feed it the information it needs.
Me: I corrected Claude and gave it an example.
Boss: Cool, did it learn?
Me: No, it made the same mistake again but it worded it differently because it can't learn anything; it just changes what the most likely text is based on those new sentences.
***
And side note: this was not a small error. Claude recently integrated with Klaviyo (and email marketing platform) and it suggested getting rid of our email leads based on a property that it mistakenly thought would eliminate people who weren't clicking our emails, thereby saving us from emailing the wrong people.
In reality, it would have gotten rid of every new lead we acquired every day.
A multi-million dollar mistake by an AI that was supposedly designed to work with this exact tool... because it simply cannot fucking learn anything by nature.