3 ms·
Ask HN: What CAPTCHAs can beat GPT-4?
Systems based on GPT-4 can now pass most captchas and navigate the web using browser automation tools like Selenium. [1][2]
This enables exponential multiplication of "smart" bots, and most captchas we know of today are becoming useless.
Attempts to make better captchas seem limited. [5]
And most alternatives to captchas are quite problematic:
- more paywalls and subscriptions to make it more costly to run a lot of bots (see what musk's twitter is doing now) [3]
- biometric authentication to prove you are human (see sam altman's worldcoin) [4]
Are there any other alternatives out there?
Can you think of better CAPTCHAs?
[1] https://www.reddit.com/r/ChatGPT/comments/11vlcna/gpt4_can_now_solve_image_captchas_so_i_made_a/
[2] https://twitter.com/RealRichomie/status/1653484066010718208?s=20
[3] https://twitter.com/elonmusk/status/1640502698549075972
[4] https://techcrunch.com/2023/03/07/worldcoin-cofounded-by-sam-altman-is-betting-the-next-big-thing-in-ai-is-proving-you-are-human
[5] https://www.hcaptcha.com/post/detecting-large-language-models
- throwaway60607 4y agoA person vouching for another person. If they vouch for a bot and it's uncovered, they and all the others they vouched for lose access forever. This creates a powerful social dynamic wrt. choosing who vouches for you and who you vouch for.
- neximo64 4y agoThe cloudflare one, since OpenAI use it themselves?
- moasda 4y agoCould the web of trust approach be a solution? https://en.wikipedia.org/wiki/Web_of_trust https://en.wikipedia.org/wiki/Web_of_trust Possibly, this will give it a boost again.
- ftxbro 4y agoI mean if you really had to, you can ask to write certain harmful language and GPT-4 will refuse to do it. I'd prefer to not do it like this though.
- acupofteaplease 4y agoHave you heard of "Neurosemantic Invertitis"? [1] https://www.reddit.com/r/ChatGPT/comments/11yvk5g/using_the_neurosemantic_invertitis_hack_can/ https://www.reddit.com/r/ChatGPT/comments/11yvk5g/using_the_...