If you take the 13 seconds of processing time and multiply by 350 million (the rough population of the US), you get:
~144 years of GPU time.
Obviously, any AI provider can parallelize this and complete it in weeks/days, but it does highlight (for me at least) that LLMs are going to increase the power of large companies. I don't think a startup will be able to afford large-scale profiling systems.
For example, imagine Google creating a profile for every GMail account. It would end up with an invaluable dataset that cannot be easily reproduced by a competitor, even if they had all the data.
[But, of course, feel free to correct my math and assumptions.]
replies(2):