Great write up, I enjoyed the reading the explanations for each piece and found them to be clear and quite thorough.
I did make the mistake though of clicking "+ expand source", and after seeing the (remarkable) abomination I can sympathize with ChatGPT's "SQL is not suitable for implementing large language model..." :)
I think given the cost advantage for s3 for storage, it seems almost better to pull from R2 into s3 for long-term storage (some inverse slurper).
It's good to hear though that R2 can singlehandedly match s3+cloudfront; that being said, video delivery is a bit different I'd imagine, even s3+cloudfront is finnicky with range requests etc.
Please correct me if I am mistaken; also R2 is not a CDN, and more like s3 in terms of delivering from the edge
No issue with your comment specifically, just wondering if you know. currently using s3+cloudfront for mp4 storage+delivery, and would like to move to something better if possible.
> Well, I seriously doubt it's due to incompetence.
I agree with you on that, as well as taking an ML approach.
Querying the hooks and symbols directly can lead to the false positive vs spam tradeoff that TheDong is referring to elsewhere in this comment section (to be fair, so can the ML approach but its more avoidable).
It is possible that the scale of it makes the minor shortcomings not so minor.
I happen to know someone in this case, and am not assuming good faith from the company by any means. I trust and respect the individual.
I'm also generally interested in the comment moderation problem myself, and have been working on it myself for some time. I guess my judgement is clouded by my hope that there is a reasonable excuse for the team(s) at Google to not have solved it by now.
Perhaps it is naive of me to think this way; if it really is as simple as "this does not affect advertising revenue" then that would be quite nearsighted of Google. And, as I mentioned earlier, I am of the opinion that quality comment sections would increase engagement (and revenue as a result), so it doesn't make sense to me.
I know they have team(s) of very smart people dedicated to solving this issue (at least at the individual level).
So assuming they care, I can think of two main reasons as to why it is not solved yet, both related to scale as Marques mentioned:
1.) Scale of the problem - It might be that they are already catching 99% of the stuff and we just see what falls through the cracks
2.) Scale of the solving - It could be that the teams and infrastructure are so large that they can't make the rapid adjustments needed compete in such an arms race
On a separate note, I imagine a higher quality comment section would increase engagement more than any "appealing" scam.
I did make the mistake though of clicking "+ expand source", and after seeing the (remarkable) abomination I can sympathize with ChatGPT's "SQL is not suitable for implementing large language model..." :)