Nick when are you going to do AI on CBreaker?
People are already cracking 100% Google Recaptcha, After reading another post on Medium how they cracked with python and tensorflow Really Simple Captcha (WordPress CSM) in 15 minutes with 100% solve rate.
They even have here their script + 9999 Solved Captcha of Really Simple Captcha (a captcha so simple, I just dont understand why CB has a success rate of like 2% on it).
https://s3-us-west-2.amazonaws.com/mlif-example-code/solving_captchas_code_examples.zip
So after trying to improve this captcha in CB and "brute force" it I came to realization why CB is so bad compared to these other networks.
CB can be hugely improved if it has simple logics build into it.
Brute Forcing it is just plain and simple trying to brute force all possible filters. But WITHOUT LOGICS.
you can simply waste an hour on 2 on filters that you already know not gonna work.
But what does CB do? Still Brute forcing it with same filters just adding more % or scaling it even further.
Even though a human can see, this is not going to work out.
So my question to @nick is, Why don't you add simple logics to the CB SDK?
Like if it sees that it is failing with a filter, why try another one after a x amount of retries.
OR..
sometimes what i have seen is with certain filters it gets like a 30% success rate.
But then moves on to a totally different filter which gets you like 50% succes rate. (with totally different filters)
Why not build a system in CB that it first going to MASSIVELY Brute Force (as quick as possible) with all filters that are similar. And saves that with filter 1,2,3,4,5,6,7 it has between 10-50% success rate.
and then second step in system could be trying a combination of filter 1,2,3,4,5,6,7 to even futher increase success rate%
After it has done that, save these filters and add on top new ones (again step 1 but with saved filters).
Right now i feel like the system of CB is... DUMB.. I dont know how else to say it.
Without any logics (doesnt have to be AI logics) but right now CB SDK is plain and simple... A Stupid machine...
They even have here their script + 9999 Solved Captcha of Really Simple Captcha (a captcha so simple, I just dont understand why CB has a success rate of like 2% on it).
https://s3-us-west-2.amazonaws.com/mlif-example-code/solving_captchas_code_examples.zip
So after trying to improve this captcha in CB and "brute force" it I came to realization why CB is so bad compared to these other networks.
CB can be hugely improved if it has simple logics build into it.
Brute Forcing it is just plain and simple trying to brute force all possible filters. But WITHOUT LOGICS.
you can simply waste an hour on 2 on filters that you already know not gonna work.
But what does CB do? Still Brute forcing it with same filters just adding more % or scaling it even further.
Even though a human can see, this is not going to work out.
So my question to @nick is, Why don't you add simple logics to the CB SDK?
Like if it sees that it is failing with a filter, why try another one after a x amount of retries.
OR..
sometimes what i have seen is with certain filters it gets like a 30% success rate.
But then moves on to a totally different filter which gets you like 50% succes rate. (with totally different filters)
Why not build a system in CB that it first going to MASSIVELY Brute Force (as quick as possible) with all filters that are similar. And saves that with filter 1,2,3,4,5,6,7 it has between 10-50% success rate.
and then second step in system could be trying a combination of filter 1,2,3,4,5,6,7 to even futher increase success rate%
After it has done that, save these filters and add on top new ones (again step 1 but with saved filters).
Right now i feel like the system of CB is... DUMB.. I dont know how else to say it.
Without any logics (doesnt have to be AI logics) but right now CB SDK is plain and simple... A Stupid machine...
Comments
like sometimes it solves 56% rate (which is highest) with filter
-scale
-threshold
And then it is going to try something totally different...
And i hit another success rate with it (example) 54% but with filters ;
-negate
-remove object
So what i mean with "DUMB" program, is that the only logic it has right now is when i stop the brute forcing system, it will only SAVE me the results of -scale and -threshold because it had the highest success rate and totally FORGETTING about that -negate + -remove object had also a DECENT success rate.
So there is no Logics in this application to even try out a combination of all 4 filters?
No, it just goes one by one, trying all of them, even if it is failing, still going to do it.
this is the reason why (when you brute force it) you see a "time to complete" of like 15 hours or so... Because it has no logics... Its just stupidly going through all of them one by one.
And the SAD part is, this can be reduced to MINUTES!!!
Im not joking around here...
If the "Brute Force" has simple logics of steps.
Step 1 : Scan fast through all filters and collect only filters with (ANY) form of success rate.
Step 2 : Take only those filters with had (ANY) results with the filters and try to optimize these filters
Step 3 : Try a combination of Filters that had (ANY) success rate
Step 4 : Take the highest success rate from those combined filters, and add on top more filters in the SAME STEPS 1 > 2 > 3 > 4 (but with the saved filters of previous steps)
DONE!
You now saved me 25+ hours of Brute Forcing Stupidly with all filters.
And you just made yourself a Nice Tool where everyone can now Brute Force smartly, improve the network, share their improved results, and lower our costs of 2captcha while you bank more cash on selling CBreaker
It has the ability to save more than one filterset when doing brute-force. It asks you about it when starting to use all free slots or not.
If you want to, it will collect as many as there are that make sense...it will however skip those who are not making sense as another set offers better results or already answers at least something in a way that next set will never get queried.
Keep in mind that CB does NOT know what answer is correct in a real scenario, just in SDK it knows it and the sorting of what filter is best is performed on the way to get the best success rate.
Your recaptcha complain with 2% success rate is obsolete anyway as there is only recaptchav2 now. The old one has been discontinued. And the filter "-segment" does a really good job. Just not with merged chars (-segmentV is working partly here). But I guess you will not find any easy way to segment that with published source.
Definitely different. I was reading about it here:
http s://developers.gppgle.com/recaptcha/docs/v3
More complex to set up. Seems like every site would have to be tweaked...