I doubt that that's very high up on their list of priorities.
> Such clients use a lot of bandwidth
How many AI companies are there really? Realistically, we're talking about a handful of additional downloads per video here at most.
> don't generate ad revenue
That's the real problem: Youtube is in the business of selling ads or Youtube Premium to humans. Anything that lets humans bypass both is going to be at odds with their goals.
But impressions by LLMs aren't (yet?) useful to advertisers, so it wouldn't make any difference either.
A lot of videos on YT are never watched and scrapers can run at high speed. At one point companies scraping web search for SEO reasons consumed a whole datacenter's worth of capacity, so scraping can be a surprisingly serious business (I used to work there).
The tech they're using to steadily lock down Youtube isn't new, so what's changed? The obvious answer would be AI.
Hm, good point. I wasn’t considering the long tail of never-watched videos – if the median number of views is very low it could still matter.
> The tech they're using to steadily lock down Youtube isn't new, so what's changed?
Time has passed and DRM is more ubiquitous on the web these days, and I suspect that the number of people using ad blockers might also be rising, the more ads Youtube is getting.
I don’t doubt that AI might have some effect on it, but I’d be surprised if that was the primary motivation.
Yeah. Remember also that cache misses are much more expensive to serve than cache hits for YT because it falls back to their core datacenters instead of being served from their edge CDN. Scrapers are nothing but cache misses.
> Such clients use a lot of bandwidth
How many AI companies are there really? Realistically, we're talking about a handful of additional downloads per video here at most.
> don't generate ad revenue
That's the real problem: Youtube is in the business of selling ads or Youtube Premium to humans. Anything that lets humans bypass both is going to be at odds with their goals.
But impressions by LLMs aren't (yet?) useful to advertisers, so it wouldn't make any difference either.