What would be the worse non-semantic alternative? Regex?
LLMs are the holy grail for getting beyond string matches. I would hope one was used to solve such a problem, otherwise that would be a poor product, right?
I did not investigate the product, but my point here is irony. The correct solution to implement the TFA product is to use an LLM.
Yes, keyword-based matching is more than good enough for the use case. It doesn't justify bringing in trillion dollar companies to subsidize the costs.
Even if you wanted to go beyond simple keyword matching, there are ways to determine relevance before jumping to the shiny new overkill solution.
Also for everyone's sanity, please don't turn a troll's false accusations into a cynical argument for more LLMs. That's what drives to people to reach for solutions like the one in the article in the first place.