The TikTok ban and Donald Trump's rise to power show how fragile our social media accounts are. We must normalize and invest in decentralized social media.
Captchas don’t do shit and have actually been training for computer vision for probably over a decade at this point.
Also: Any “simple test” is fixed in the next version. It is similar to how people still insist “AI can’t do feet” (much like rob liefeld). That was fixed pretty quick it is just that much of the freeware out there is using very outdated models.
I’m talking text only, and there are some fundamental limitations in the way current and near future LLMs handle certain questions. They don’t “see” characters in inputs, they see words which get tokenized to their own internal vocabulary, hence any questions along the lines of “How many Ms are in Lemmy” is challenging even for advanced, fine tuned models. It’s honestly way better than image captchas.
They can also be tripped up if you simulate a repetition loop. They will either give a incorrect answer to try and continue the loop, or if their sampling is overturned, give incorrect answers avoiding instances where the loop is the correct answer.
They don’t “see” characters in inputs, they see words which get tokenized to their own internal vocabulary, hence any questions along the lines of “How many Ms are in Lemmy” is challenging even for advanced, fine tuned models.
And that is solved just by keeping a non-processed version of the query (or one passed through a different grammar to preserve character counts and typos). It is not a priority because there are no meaningful queries where that matters other than a “gotcha” but you can be sure that will be bolted on if it becomes a problem.
Again, anything this trivial is just a case of a poor training set or an easily bolted on “fix” for something that didn’t have any commercial value outside of getting past simple filters.
Sort of like how we saw captchas go from “type the third letter in the word ‘poop’” to nigh unreadable color blindness tests to just processing computer vision for “self driving” cars.
They can also be tripped up if you simulate a repetition loop.
If you make someone answer multiple questions just to shitpost they are going to go elsewhere. People are terrified of lemmy because there are different instances for crying out loud.
You are also giving people WAY more credit than they deserve.
Well, that’s kind of intuitively true in perpetuity
An effective gate for AI becomes a focus of optimisation
Any effective gate with a motivation to pass will become ineffective after a time, on some level it’s ultimately the classic “gotta be right every time Vs gotta be right once” dichotomy—certainty doesn’t exist.
Anyway proof of work scales horrendously, and spammers will always beat out legitimate users of that even holds. I think Tor is a different situation, where the financial incentives are aligned differently.
Captchas don’t do shit and have actually been training for computer vision for probably over a decade at this point.
Also: Any “simple test” is fixed in the next version. It is similar to how people still insist “AI can’t do feet” (much like rob liefeld). That was fixed pretty quick it is just that much of the freeware out there is using very outdated models.
I’m talking text only, and there are some fundamental limitations in the way current and near future LLMs handle certain questions. They don’t “see” characters in inputs, they see words which get tokenized to their own internal vocabulary, hence any questions along the lines of “How many Ms are in Lemmy” is challenging even for advanced, fine tuned models. It’s honestly way better than image captchas.
They can also be tripped up if you simulate a repetition loop. They will either give a incorrect answer to try and continue the loop, or if their sampling is overturned, give incorrect answers avoiding instances where the loop is the correct answer.
And that is solved just by keeping a non-processed version of the query (or one passed through a different grammar to preserve character counts and typos). It is not a priority because there are no meaningful queries where that matters other than a “gotcha” but you can be sure that will be bolted on if it becomes a problem.
Again, anything this trivial is just a case of a poor training set or an easily bolted on “fix” for something that didn’t have any commercial value outside of getting past simple filters.
Sort of like how we saw captchas go from “type the third letter in the word ‘poop’” to nigh unreadable color blindness tests to just processing computer vision for “self driving” cars.
If you make someone answer multiple questions just to shitpost they are going to go elsewhere. People are terrified of lemmy because there are different instances for crying out loud.
You are also giving people WAY more credit than they deserve.
Well, that’s kind of intuitively true in perpetuity
An effective gate for AI becomes a focus of optimisation
Any effective gate with a motivation to pass will become ineffective after a time, on some level it’s ultimately the classic “gotta be right every time Vs gotta be right once” dichotomy—certainty doesn’t exist.
@NuXCOM_90Percent @brucethemoose would some kind of proof of work help solve this? Ifaik its workingnon tor
Somehow I didn’t get pinged for this?
Anyway proof of work scales horrendously, and spammers will always beat out legitimate users of that even holds. I think Tor is a different situation, where the financial incentives are aligned differently.
But this is not my area of expertise.