FMS makes use of audio captchas. A response to abuse could be to move in
that direction as well.

On Jul 20, 2016 6:19 PM, <[email protected]> wrote:

> > On 20-07-16 17:05, [email protected] wrote:
> >> Google research declares the simple text captcha dead. They remain
> >> difficult for human beings, but machine learning can crack them 99.9% of
> >> the time.
> >
> > While probably true, no captcha abuse on WoT has been documented so far.
> >
> >> I propose a better captcha for WOT that works like:
> >
> > Thank you for considering how WoT can be improved. Your proposal is not
> > immediately clear to me; I'll fire some questions and state some of my
> > doubts to get a better understanding.
> >
> >> - The challenge to be solved: Add the numbers presented in the puzzle
> >> series that ONLY show trees.
> >
> > I don't see what benefit having to add numbers brings over recognizing
> > characters. Digits are easier to recognize for computers (for there are
> > only 10 of them, compared to our 26 character alphabet), and addition is
> > a trivial task for computers as well.
> >
> >> - It presents 5-10 puzzles. The legit puzzles randomly sprinkled
> >> throughout.
> >
> > What do you mean by randomly sprinkled legit puzzles? All puzzles in WoT
> > are supposed to be legit. What purpose would a non-legit puzzle have? Or
> > do you envision a series of multiple "puzzles" within a single
> > captcha-like image?
> >
> >> - What puzzles look like: A number is drawn across an image backdrop of
> >> trees and other objects.
> >>
> >>
> >> That should raise the bar and not be hard to switch to.
> >
> > In my humble opinion, this statement is far from the truth.
> >
> > Consider a finite set of "tree" and "non-tree" images that is to be
> > shipped with WoT. This reduces your proposal to a captcha with an
> > alphabet extended to the cardinality of the set of images, however
> > operating on image similarity (and known classification information on
> > whether the image is a tree or not) instead of character similarity.
> >
> > The strength of captcha alternatives that are based on identification of
> > image contents, relies on the assumption that the image set (in other
> > words, the classification data) is kept private. This cannot be done for
> > WoT (the attacker could always download the source after all!), unless
> > each user would have to supply their own source of captcha images —
> > which would make setup unnecessarily hard.
> >
> > Kind regards,
> > Bert
> >
> > _______________________________________________
> > Devl mailing list
> > [email protected]
> > https://emu.freenetproject.org/cgi-bin/mailman/listinfo/devl
>
> I got nothin' :) The idea was poorly thought out. Thanks for your time
> _______________________________________________
> Devl mailing list
> [email protected]
> https://emu.freenetproject.org/cgi-bin/mailman/listinfo/devl
_______________________________________________
Devl mailing list
[email protected]
https://emu.freenetproject.org/cgi-bin/mailman/listinfo/devl

Reply via email to