Given my experience with ads today is that they frequently get my gender, language, nationality, and country of residence wrong (and in mutually incompatible ways), I suspect that the famous pregnancy story was dumb luck rather than a reflection of quality.

But yes, the Al Capone solution seems like it would work.

They don’t have to be right all the time, just enough to make ROI positive.

I realize this is the prevailing wisdom but it is short sighted and simplistic because it ignores opportunity cost.

Could you expand on that?

Simply being positive does not mean it is ideal. Accepting the first approach that delivers a positive result could mean you miss out on other even better opportunities.

For their own needs, yes. For the purposes of satisfying the goals of government regulations, no.

I don't mind them having a positive ROI. I do mind them spying on me and my children, even if they draw incorrect conclusions about what ads interest us.

More on mechanism (and why specific mechanisms should not be legislated) here:

<https://news.ycombinator.com/item?id=42075146>

> SocMed / AdTech cos. can already ID and target users with precision.

Not reliably so, and not people who are deliberately trying to fake the system.

Looking at the ads FB are showing me, they are not well targetted at all. I can see an add about AI in education that seems to be aimed at teachers (not a teacher), one from a charity aimed at people with spare rooms that could be used to accommodate human trafficking survivors (I do not have a spare room), a summer school for kids who might do a history related subject at university (that one is well targetted), one advertising admin jobs at British intelligence agencies (I only just meet security requirements and I do not have relevant experience). I get shown multiple ads aimed at Muslims (I am not a Muslim). In the past I have been shown ads for Judaism GCSE courses (not Jewish either, and my kids did not do any religion GCSEs).

As a counter to this: advertisers not knowing how to use the targeting options available to them != the platform can't target well.

Saying that though, it does seem like as soon as I click one advert (either out of curiosity, or to waste obvious scam advertisers budget), I'm suddenly inundated with similar adverts.

Worse I've clicked an ad, bought a product and still seen adverts for the same product from the vendor after the fact. I'm pretty sure there are ways for vendors to preclude such wasteful ad spend.

> As a counter to this: advertisers not knowing how to use the targeting options available to them != the platform can't target well.

Also, it's not like the platform is going to give advertisers access to its full capabilities. For one, they don't need to - they only need to sell the minimum it takes to keep the advertisers on board. Two, they wouldn't want some clever advertisers to extract juicy data indirectly and cut out the middleman.

Wut?

Better capabilities for customers, they reach more customers, more revenue = they may increase ad spending because they see it working

Yes, but you want to minimize the "better" part so you get the extra revenue without losing your competitive advantage.

it's also helpful to extract all the value at the current cost before incurring more costs.

> some clever advertisers to extract juicy data indirectly and cut out the middleman

What am I missing here? Don't you still need the platform to actually advertise on? How could I extract data and use that to advertise to someone?

Rather than tax, I would just like to see data on how many Ads kids are being shown in a day. And whether its increasing year on year.

¿Por que no los dos?

And yet Netflix, Amazon, etc recommend movies and shows I’ve watched in the last 6 months. They literally know who I am and what I’ve watched and can’t/won’t build a basic recommendation engine that includes “don’t recommend things watched in the last X months”.

Maybe there’s a reason they want me to rewatch things, but I find it extremely annoying to have the first rows of recommendations be things I watched recently.

That is all but certainly more a sign of enshittification than of technical inability to implement that function.

This might make sense given the priors. e.g. someone who's hasn't watched Star Wars in the past 6 months might be less likely to watch and enjoy Star Wars than someone who's already watched Star Wars in the past 6 months.