@jjuddpi ? I have no vested interest in proving it. Their tool is a weak classifier that is already being outdone by frontier models. Their “bug bounties” only serve to make the product look legitimate because they need the PR. They’d just move the goalpost like they do anyway
@jakehalloran1@BrandonLive Wow what a ground breaking observation. It is very easy. If you haven’t been able to make false positives you aren’t trying hard enough. It is particularly noticeable on anything that contains lists
@jjuddpi Why would I do the work for pangram? Their bounties are just memes for their shills like you. They exist to market the product and nothing more
@jakehalloran1@BrandonLive It is just as easy to create false positives. I have tested it EXTENSIVELY (see how meaningless that is). Victor Hugo, the whale poem, etc etc. the list goes on of actual false positives they just patch and hope you don’t notice.
@basedpolymer@kgrotov Victor Hugo ~1 month ago is a well known example. You can also find other examples of people posting how they went from 100 ai to 100 human all over twitter. I have seen it mark things in my own personal prior writing too.
@basedpolymer@kgrotov This substack opinion piece only shows pangram has significant limitations/caveats. Experiment with your own writing. I’ve found it very easy to go from 100% ai generated or not with just a few swaps. It’s also easy to find things it will always say are ai, despite being human
@norpadon This is the kind of mindset that sets our kids up for failure. They are the next generation and whether or not they have “meaningful experiences” doesn’t give us a right to view them as more expendable than adults. This is a weird thing to call “based”