i tend not to comment on CS shit on tumblr because almost nobody here is nuanced-tech-literate in a way that doesnāt get 18,000 reactionaries in your notes, but iām not lying when I say this newest fucking AI imagery model ānano banana proā should be illegal lmao. not even as a āskeptic,ā not as some rube fighting the war on copyright on the side of the copyright, I mean it should be outwardly illegal to synthesise images this realistic on grounds of potential dis&misinformation:
the singularity straight up just happened on some rainy Wednesday in November. not because we hit AGI, but because you can now hop on googleās newest website and generate a near perfect, almost indistinguishable image of anything you want within a matter of minutes. from here on out again, itās going to get even harder to discern GAN imagery from real imagery. we had close calls with earlier models but they have deliberately trained away from specific giveaways involving texture and lighting with this model.
and if youāre thinking about being smart at me over this - donāt! I know it takes prompt engineering and luck to get results this good but the fact we have achieved it in non-lab settings without fine tuned training for specific outputs (these specific outputs are based on nbp & a dataset from Instagram models/influencers specifically) is cause enough for concern.
in fact this is so insanely irresponsible iām actually scratching my head at why the fuck this is being deployed for any public use at all. this shit is a fantastic achievement in academia but in the hands of not only the public but the governments, military, legal system ā¦
iām not gonna say itās doomsday, iām not a doomer, but this is still like. ostensibly horrific. either get good at IDing fake content in the next few days as more comes out of this model, or for the love of god, do not trust the voracity of anything you see post Nov ā25. we built a tool to grant us the fruit of knowledge and it was too much for our psyches to bear .god have mercy. lmao
