I suppose we disagree on whether LLMs as a method are currently "well understood to work better (than not using LLMs in any capacity)" or not.

That's not really what you claimed in the first part though. Your argument there was more along the lines of "regardless of how well your current tool works, you shouldn't use it if there's a likelihood a more efficient tool will come along in the future".

I mean, doesn't the below argument apply to like, buying a computer? According to Moore's law one has/had a "late mover's advantage" by where waiting to buy for two years doubled your value for money. So why do you own a computer already, aren't you better of saving your money?

> There's really a weird sort of late mover's advantage to just holding off and saving your money unless you expect your profit to grow faster than AI becomes more powerful.