Hacker News new | ask | show | jobs
by mrspeaker 2750 days ago
I'm sure that by now the ad networks must have figured out that I compulsively lie to them and take that into consideration. Youtube just can't believe that I've mysteriously never heard of a single product they ask about in their surveys... "Wow, this guy has never heard of Starbucks either?!"

Lying to CAPTCHA is getting tougher these days - it takes me a few minutes now before they let me through even though I've identified a plain piece of road as a sign. It's a pain to sit through until they let me go, but I'M NOT YOUR FREE TRAINING DATA!

3 comments

That is part of why I switched away from Google & containerized Youtube in Firefox, its really easy to end up in CAPTCHA hell it seems. Qwant and DuckDuckGo are sufficient, and actually better when searching specific item names or part #, its pretty impressive how shit Google's results have become as of late for this use case IMO.
The only reason nowadays to use Google search is verbatim error messages for exotic software products.

There it still seems to outsmart DuckDuckGo. Else than that I haver zero reason to look back. As a matter of fact. That little box, containing basics for questions like "shell date operations" is usually sufficient and if not you have a link to (usually) stackoverflow. I really like the concept.

On a side note why does Google offer a captcha service to the public, then for their services use an incredibly shitty one. It takes me 4-5 tries to decrypt the random squiggles on their sites. I don't mind decoding a bit of text or an address, but lately even their public captcha has gone rapidly down hill.
I was recently thinking, I'm not entirely sure if they would still use those photo-CAPTCHA's for ML training data any more. The classification "puzzle" that is offered seems like something deep learning is already capable of today? It seems like such "toy data", considering the images and the task given.

I could be completely wrong though. Is there anyone up to date on modern ML capabilities to comment on whether this data is useful and what for? I used to think it made sense, especially with ReCAPTCHA (digitizing books) but it just doesn't seem that valuable any more?

Getting captcha correct on google is actually difficult. For some reason they use a different one than they provide for third parties.

As for survey/reviews I routinely one star apps that nag me to leave a review. Some apps have two options "review now" or "remind me later" . Those apps get reviewed, "app is great, but won't stop fucking asking me for a review even though I already have."