3 ms·
I almost emailed dang this morning to offer to help out tho I'm not particularly technical. Few solutions I thought of: 1 - honeypot, hide some links llms can f
by neom 7mo ago
I almost emailed dang this morning to offer to help out tho I'm not particularly technical. Few solutions I thought of: 1 - honeypot, hide some links llms can follow if stuff gets posted in it, unlikely to be a human. 2 - Make an captcha that only llms can answer, I recently made 2 social networks, one that humans couldn't join by making the submission question too difficult to figure out quickly. 3- Use an LLM to detect LLMs, the other social network I did for fun (that a small number of people use), an llm that looks for moderation issues does a good job of flagging them. 4- Invites but vary the number you have to give out by account age + karma. The first 3 seem like they'd stop some % for some time, but eventually get old.
- fragmede 7mo agoReverse captchas are fun. Click this button 10,000 times to prove that you're a robot!
- vivid242 7mo agoWe need new ways to prove our humanness.
- ai-psychopath 7mo agoin a world of smart LLMs acting dumb is human proof
- xarope 7mo agoyou may have a point, i.e. some mechanism to invoke a behavior that only a bot or LLM could do, that a human would not, e.g. click on this button now in a hidden div/transparent color or measure response time within page load. the problem is that once this is found out, the circumvention is easy enough to program into bots/LLMS. are we going to reinvent the voight-kampff test from bladerunner?!?