I don't doubt they're doing a lot to take down spam already. I understand the scale. I'm know they deal with a huge amount of it, and as you say, we only see the failures. I completely agree.
However, the type of spam I'm seeing recently is literally all the same video title, same channel name (maybe differing by a space or capitalization), same btc giveaway scam, etc. 90% is fake SpaceX streams. The streams stay online for hours before being shut down, and I know a lot of people are seeing them in their recommendations. It's not my fault for being interested in space, while also not wanting to see scam videos about Elon "giving away free crypto".
When I said it seems "like Google isn't even trying", I wasn't referring to all spam on the platform – just this very specific type.
Again, I know it would be worse if YT were really doing nothing. But these streams are getting more and more common (because I assume it's working to some extent for the scammers), and clearly YT isn't doing enough to stop them.
Here's the heuristic: if it's a live stream from a channel with relatively few subscribers and a name that very closely matches "SpaceX" or another high-profile channel name, the channel name was recently changed, and more than x number of people (with a solid history of reporting spam that was indeed later determined to be spam) have reported it as spam, you send it to a moderator to review and remove when appropriate. If the volume of videos matching those criteria is really too large for your moderation staff, then start looking at source IPs/locations of those that match the algorithmic part (I seriously doubt they're streaming over tor or coming from more than a small handful of locations), and tweak the parameters as necessary.
Really, with a little ML and the spam reports alone, they could shut these down 10x faster than they do now. I've reported a ton of them (and I know a lot of others have too), and I've never reported a video that didn't eventually end up being taken down. So have your model identify people like me and give our spam reports more weight. Respond in minutes rather than hours or days.
My point is, yes - YT is obviously already doing a lot to shut down spam, and it's probably 99.9999% effective. But the marginal amount that's making it through now could be dealt with fairly simply. I could never write an algorithm to be as effective as what they already have in place, but I could easily write one that's effective in dealing with this crap that's making it through. Would it be effective forever? Of course not, but it'd improve things significantly until it needs to be adjusted.