a year ago, i attempted building software that detects ai-generated images i quickly realized it was becoming a dog chasing its own tail newer, more capable models kept emerging, and i always had to keep up with finding weaknesses we'll soon approach an era where the difference between agentic and human outputs will be +epsilon we have a new, very hard problem: no matter how smart the software is, detection will eventually fail so if we want to solve this problem, we must use something that is not purely algorithmic i wonder, how many real use cases need this problem solved today, and how many will there be in 5-10 years from now? once the biggest models get distilled, uncensored, and generally available/open-sourced, no one will be able to verify the authenticity of media through content alone, not even the same labs that built those models in the first place i also wonder if achieving this is part of the agenda granting the ability to shape reality and present it to the masses as if it really happened, erasing the thin line between real and unreal regardless of the danger, there will come a day when proof of provenance (human origin) is required, and that day might come sooner rather than later