Do the captcha words get more difficult over time?

Question

I read every page on the website, but it doesn't say anything about what I want to know, except this, which is peripherally related:

Our service also includes IP address filtering and detection. If we determine that a given IP address is successfully solving too many CAPTCHAs in a certain period of time, the address is immediately flagged for review.

So presumably when I go on one of my retagging sprees, the system sits up and pays attention. However, even days later of little activity, I've been getting more difficult-to-type words, regardless of the frequency I hit the captcha.

I started with only easy words like 'with', 'and', etc., but now I'd say half the time it's giving me long/difficult-to-type words like 'congressional' (that's the only one I remember from my session a few minutes ago).

Am I imagining this, or are the words actually supposed to be more difficult as time goes on? I'd be particularly interested to hear from some high-rep users who have done a lot of manual retagging.

(Yes, this is a bit of a whine post, but this, combined with the low captcha threshold, is making it more time-consuming and annoying for me to do retagging, which really detracts from the job I want to do.)

Edit: Here is a sampling of the captchas I got during my recent retagging session just now. These are pretty easy ones compared to some I've seen.

Captchas http://jonseigel.info/images/StackOverflow/Captchas.gif

Answer 1

In theory, at least, it will get harder.

The recaptcha widget used here is based on a project that scans/digitizes books. The captcha we see are created from words that the ocr scanner was unable to identify. So, it works well as a bot preventative because we already know that the one of best ocr engines money can buy is unable to read it.

However, these engines are improving all the time. As the engines improve, the difficulty in the words they are unable to read should increase as well. But in one sense this is a good thing, as it means we automatically stay one step ahead of the hackers. It's a similar concept to the BCrypt password hasher, that we scale up the problem automatically as the attack sophistication increases.

On the other hand, it's also only a matter of time until the machines are better than humans. But at least for the moment recaptcha is the state of the art.

I also don't think it's flagging you for more difficult captchas right now. I doubt that it scores them, though that certainly is a possibility. And if you think "congressional" is hard, remember that these words are pulled from real books. Real books don't always have nice clean words. Sometimes you see 1/2 of a hyphenated word. Or a number. Or punctuation. Or accents, foreign characters, digits, foreign words...

Answer 2

No, but as you get older it gets more and more difficult to remember anything beyond what you were doing in the last five minutes. Coupled with the degradation in eyesight once you hit 40, it's a double whammy when trying to solve a CAPTCHA. Perhaps this is a sign that you've started the inevitable downhill slide though, according to your profile, you seem a little young for that.

Answer 3

Now, in the amazing world of the future, we know the answer is a definite "yes".

Here's one I dug up from May 2010:

August 2012:

The first one was complained about because it has the Phi character, but other than that it's clearly more legible that the current versions. The letters are all the same width, with virtually no overlap.

Futhermore, the 2010 example is actually much harder than most from that period--I remember most of them being two normal words, whereas the 2012 example is par for the course. To sum up, ReCAPTCHA is succeeding in its primary goal: improving OCR software.

Answer 4

No, they are not more difficult.

With that being said, even if they were (again they are not), only one of the two words could be "more difficult", because for each captcha the server only knows the answer for one of the two words. They are read from actual books and since computers can't read them - that's the whole point - the server has no way to know how "difficult" it is.

(A little tip; if one of the two words is illegible or very difficult then it's most likely the not-yet-discovered word and just typing anything - for example, a - will work almost every time).