[saurik]

Yeah :/.

It should be "this person bought a vacuum cleaner today, so let's wait two months and start showing them ads for vacuum cleaner bags" or "maybe they are buying equipment for a new apartment, so maybe they would also like a toaster oven and a lamp".

Like, imagine you had an actual salesman--a good one, as opposed to the "targeted salesman" from the video--who you bought all your stuff through and who was excited to make commissions off of you. They would never waste your time or attention or trust asking you to buy a second vacuum cleaner.

If someone buys a treadmill, they probably don't want a second one, but maybe they now want some free weights. The current algorithms are just so stupid... like, I almost appreciate the premise that if you had an "actual AI" helping find you stuff to buy that could be fun, but what we have is just wasting everyone's time.

[visarga]

This seems like sloppy ML, there are sub-fields that deal with sequences and time series and would be great at modelling such complex interactions. Language models do something like that.

My bet is that all the fancy ML is useless because the sellers are not adapted to the buyers. They want to sell X and buyers need/want to buy Y. If they used proper targeting they would reach a much smaller audience.

So they have an order to show 1000 crappy flash drive ads, but you only want to buy other things - what are they going to do? Skip you? No, that's money lost. They will show you the ad even if it is ineffective. Greed explains the bad targeting.