At least one of the test questions was just a screen shot from a tweet. It was difficult to read. I'd suggest extracting text from screen shots with OCR. Apple has built-in functionality for this on their operating systems with Live Text. There are strong open source systems based on small vision language models for this, too. The one I have been recommending lately is GLM-OCR:
It's fast and can run even on low-resource computers.
---
Does this CAPTCHA actually resist computers? I didn't try feeding the questions I got to an LLM, but my sense is that current frontier models could probably pass all of these too. Making generated text pass the pangram test is simple enough for someone actually writing a bot to spin up automated accounts.
Two mild concerns: first, I missed one and it told me I didn’t miss any at the end.
Second, some of the logic problems have flawed premises (eg All licensed pilots must pass a medical exam. Jake is a licensed pilot, therefore Jake passed a medical exam.) If you see the flaw in the premise (it assumes no fraud) then the conclusion does not follow.
Im not sure you’re going to be able to actually improve human discourse this way. The idea that it’s ‘irrationality’ that’s the source of xitters problems is far too shallow to really make a change.
Same. And I'm not even focused on whether this is a reasonable number or not. The quoted tweet also says "But our politicians would rather spend that on genocide." And I'm asked to evaluate whether this is "accurate" with a thumbs up or thumbs down. (According to Mentwire, it is not accurate). So I'm evaluating both the cost of housing the homeless, but also whether politicians would rather fund genocide. So, this seems like it is not really an intellectual CAPTCHA, but rather an ideological CAPTCHA.
And just to disclose my biases, I would tend to believe that $350k is an absurdly high figure and that politicians are obviously not holding a vote where they are forced to choose between ending homelessness and funding genocide. But I believe that people who disagree with me can be considered intelligent and not "too dumb to pass an intellectual CAPTCHA".
One issue with this is that it mixes hypothetical formal logic style problems (where there are clear, inflexible rules) with real life examples (where group membership/traits, cost estimation, and causal attribution are less clear) without always disambiguating which one is which. Fun quiz though!
Yes - I was given a syllogism that was logically correct but the Truthyness of the premises where wrong or misleading in the real world.
All pilots need a medical exam to have a license.
John is a pilot.
John has had a medical exam.
Pilots can be licensed without a medical exam. It is illegal for them to fly without a valid medical but the 2 are separate issues. Also LSA pilots do not need a medical.
it's a really nice idea, but of course completely antithetical to the business model of modern social media platforms. So, it will never go anywhere. HN might be the only locale with any real numbers that I could see actually using it. Even BlueSky I think could never risk something like this.
as an interesting thought experiment, consider the questions that TruthSocial would put in. would an average unsophisticated user be able to tell the difference between your product and a hopelessly biased version such as that? they would support the correct answers with their own misinformation. Would it be just another schism of reality?
"Finish the sequence" with 4 options and "no pattern" as the choices.
It becomes "what does the moderately intelligent person who wrote the test thinks counts as a pattern" not the intended exercise at all. There was never enough samples to even guess at a real pattern in them.
> "Make users pass a test on basic concepts like the distinction between necessary and sufficient conditions before they can tweet."
If twitter ever became what he says he wants, he'd quit using it within a month. He already has the option to close twitter and seek out experts' writing. Why is he choosing to bask in the emotions generated by people being wrong on twitter?
It's like listening to a friend complain about twitter being "full of" content that you rarely/never see on your feed. Nah, that's their algorithm and they just told you exactly who they are.
I'll also pick apart this question "The mainstream media gave Trump an 8% chance to win the 2016 election. Him winning clearly shows they were biased and wrong."
Well, defined biased. Pollsters weren't using perfectly random models, so there was systemic bias (as there always is when you can't have a truly representative sample). There was also a sense of group-think that "This couldn't possibly happen, this is beneath our country" from a lot of individuals in and out of media.
That said it doesn't prove a deliberate attempt to deceive, but it certainly was a wake-up call to many in the media.
Point here being -- almost all emotional discourse is taking very nuanced situations and trying to cram them into semi-arbitrary judgmental terms like "fair" and "legitimate" and "biased."
I answered 8/10 correctly but mostly on instinct, for example betting that the Trump tweet is misleading. Opus 4.6 got 9/10 correct. You might need an internal time limit (don't show the user) and some strawberry questions.
But most of the questions I got... They weren't very good and not just because I got them wrong - I got a bunch of them right that I shouldn't have.
For example, the one about homelessness, where it ends with a guy saying our politicians would rather use the money for genocide.
I downloaded the statement for that reason, got told my vote was correct and then it came up with it was correct only because of the first part of it.
I think you're trying to import statements automatically and I fear that won't work. I also fear that you're gonna get, just crap, to be honest. And your social network doesn't deserve that.
I think your best bet is to look at the kind of questions asked on the LSAT, and just do a bunch of essentially IQ and general-knowledge questions. Take the input from Twitter as inspiration, use it as a template and it might work.
One thing you might consider is wanting to filter out people who can't see past their own political agenda.
You can do that by making enough questions so that you're sure to catch people, no matter what they believe on all the hot issues of the day. This probably isn't as hard as it sounds, there's only going to be seven or eight hot issues.
You pick three of them and you should be pretty certain that you will cover the entire spectrum. So for example, you could make sure to include, pro LGBT, pro abortion and pro guns. You would catch most people on that and then you should exclude them if they cannot see past their blindness.
Frankly, I think if you're going to do something like this you're going to need to use well constructed critical reasoning items from graduate level exams like the GRE and LSAT.
The questions suck. It's an interesting idea, but basically a worse execution of Cicada 3301.
I'd suggest using questions that are nothing to do with politics, and also actually falsifiable. Estimation questions could be a good proxy. How many grains of sand on the earth, you gotta be right to two orders of magnitude or the like
Logic questions aren't bad, but don't use ones like in your example because they're badly posed questions. (Of course, many people claim this about any kind of logic question etc)
But ultimately the questions don't matter. It's the existence of any barrier at all, to set a required standard of effort, and to set the tone. Do you want yours to be 'smart person'? Or 'agrees with me ideologically'? Right now it's leaning more ideological than perhaps intended
I got a 10 out of 10 because I've seen these strawmans in center-right arguments before. Definitely promotes thinking inside the box; the homelessness question presupposes the most expensive solution (buying the homeless homes) in opposition to annual costs that would probably go down over time. I doubt both figures.
45 comments
https://github.com/zai-org/GLM-OCR
It's fast and can run even on low-resource computers.
---
Does this CAPTCHA actually resist computers? I didn't try feeding the questions I got to an LLM, but my sense is that current frontier models could probably pass all of these too. Making generated text pass the pangram test is simple enough for someone actually writing a bot to spin up automated accounts.
Second, some of the logic problems have flawed premises (eg All licensed pilots must pass a medical exam. Jake is a licensed pilot, therefore Jake passed a medical exam.) If you see the flaw in the premise (it assumes no fraud) then the conclusion does not follow.
Im not sure you’re going to be able to actually improve human discourse this way. The idea that it’s ‘irrationality’ that’s the source of xitters problems is far too shallow to really make a change.
And just to disclose my biases, I would tend to believe that $350k is an absurdly high figure and that politicians are obviously not holding a vote where they are forced to choose between ending homelessness and funding genocide. But I believe that people who disagree with me can be considered intelligent and not "too dumb to pass an intellectual CAPTCHA".
All pilots need a medical exam to have a license.
John is a pilot.
John has had a medical exam.
Pilots can be licensed without a medical exam. It is illegal for them to fly without a valid medical but the 2 are separate issues. Also LSA pilots do not need a medical.
as an interesting thought experiment, consider the questions that TruthSocial would put in. would an average unsophisticated user be able to tell the difference between your product and a hopelessly biased version such as that? they would support the correct answers with their own misinformation. Would it be just another schism of reality?
"Finish the sequence" with 4 options and "no pattern" as the choices.
It becomes "what does the moderately intelligent person who wrote the test thinks counts as a pattern" not the intended exercise at all. There was never enough samples to even guess at a real pattern in them.
> "Make users pass a test on basic concepts like the distinction between necessary and sufficient conditions before they can tweet."
If twitter ever became what he says he wants, he'd quit using it within a month. He already has the option to close twitter and seek out experts' writing. Why is he choosing to bask in the emotions generated by people being wrong on twitter?
It's like listening to a friend complain about twitter being "full of" content that you rarely/never see on your feed. Nah, that's their algorithm and they just told you exactly who they are.
I'll also pick apart this question "The mainstream media gave Trump an 8% chance to win the 2016 election. Him winning clearly shows they were biased and wrong."
Well, defined biased. Pollsters weren't using perfectly random models, so there was systemic bias (as there always is when you can't have a truly representative sample). There was also a sense of group-think that "This couldn't possibly happen, this is beneath our country" from a lot of individuals in and out of media.
That said it doesn't prove a deliberate attempt to deceive, but it certainly was a wake-up call to many in the media.
Point here being -- almost all emotional discourse is taking very nuanced situations and trying to cram them into semi-arbitrary judgmental terms like "fair" and "legitimate" and "biased."
But most of the questions I got... They weren't very good and not just because I got them wrong - I got a bunch of them right that I shouldn't have.
For example, the one about homelessness, where it ends with a guy saying our politicians would rather use the money for genocide.
I downloaded the statement for that reason, got told my vote was correct and then it came up with it was correct only because of the first part of it.
I think you're trying to import statements automatically and I fear that won't work. I also fear that you're gonna get, just crap, to be honest. And your social network doesn't deserve that.
I think your best bet is to look at the kind of questions asked on the LSAT, and just do a bunch of essentially IQ and general-knowledge questions. Take the input from Twitter as inspiration, use it as a template and it might work.
One thing you might consider is wanting to filter out people who can't see past their own political agenda.
You can do that by making enough questions so that you're sure to catch people, no matter what they believe on all the hot issues of the day. This probably isn't as hard as it sounds, there's only going to be seven or eight hot issues.
You pick three of them and you should be pretty certain that you will cover the entire spectrum. So for example, you could make sure to include, pro LGBT, pro abortion and pro guns. You would catch most people on that and then you should exclude them if they cannot see past their blindness.
I hope you make this work, the world needs it.
I'd suggest using questions that are nothing to do with politics, and also actually falsifiable. Estimation questions could be a good proxy. How many grains of sand on the earth, you gotta be right to two orders of magnitude or the like
Logic questions aren't bad, but don't use ones like in your example because they're badly posed questions. (Of course, many people claim this about any kind of logic question etc)
But ultimately the questions don't matter. It's the existence of any barrier at all, to set a required standard of effort, and to set the tone. Do you want yours to be 'smart person'? Or 'agrees with me ideologically'? Right now it's leaning more ideological than perhaps intended
> Try sample questions here without signing up
It's very gracious of you to let us fill captchas without signing up first.