It's "obvious" because A is B -> B is A is not a thing that is actually true for the vast majority of text (or really any kind) constructions. It's only a truth of formal logic.
Logician here. We're well aware that "A=B" is not a good formalization of the colloquial/grammarical "A is B", we don't formalize it that way, and we don't regard "if A is B then B is A" is a truth of formal logic.
The people making the argument about reversal curses are not logicians, and most of them don't know anything more about formal logic than what anyone would pick up in an undergraduate "Intro to Proofs" course.
That said, the semantics of the word "is" in natural language really doesn't matter in this debate. The semantics is a red herring, if you will (while a red herring is not semantics).
After all, LLMs cannot learn "the quantity B is mathematically equal to A" from examples of "the quantity A is mathematically equal to B" either, even when the rest of the corpus clearly explains that this _is_ in fact always reversible.
It is a thing that nearly any non-technical user of LLM's would expect it to be able to do, and be surprised at it not doing. Which is a bug, if you expect it to be a service/piece of software that someone not trained on using LLM's to be able to do. The author more or less admits this in the section entitled "Model training is a dark art"
'I’ve been playing around with fine-tuning LLM models for years and still don’t have any hard and fast one-size-fits-all rules to apply. Every dataset lends itself to a specific way of training.'
But this is precisely the reason why nearly every claim about what "AI" will soon be able to do, is misguided. The failures of LLM's are exceedingly unintuitive to anyone who doesn't have a lot of experience with them (and maybe sometimes to people who do).
Spreadsheets can be used by people who don't know how spreadsheets' internals work; after a bit of training (in my experience, about ten minutes) they can get a decent intuition about how to use a spreadsheet (at least for the simple stuff). The same is true of well designed web pages, word processors, music players, etc. Even software with more complex training requirements like CAD, statistics packages, video editing, etc. will usually behave in a more-or-less intuitive fashion for a user of the appropriate target group.
LLM's fail in unexpected (and sometimes hard to spot) ways, and are being marketed as if they are a tool for the general population to use, when their training (and failure modes) are still a "dark art" even for people with years of experience in them. That is not a feature.
