Good point. So basically, they likely spent 1+ million USD letting Mythos run for hours upon hours, and in this one case, a query with sizable compute used, apparently US$100k at the end of it all, came up with something interesting. And that $1 million was just an example. What if they ran 40 such queries? That'd be $4 million bucks flushed down the toilet. Not exactly an amount people will pay for a 1/40 chance to see if an LLM can innovate. They'd rather pay teams of humans, say, $4 million in total, since they have a much better track record for the money spent.