> I suspect we're about to hear some arguments from AI-maximalists that LLaMA is sentient and that deleting it would be akin to murder - and wiping out AIs trained on stolen property is literally genocide. I don't believe that for a second.

Me thinks anthropomorphizing clocks is as silly as worrying if the boiling water in the pot feels pain. Even if it was feasible for a machine to be conscious, it would be conscious at a lower level than "software", which, to me, has come to resemble a hyper hyperbolic information layer (emerging and part/parcel of human consciousness.) (Call it an egregore if you want, but to me it seems more like a data lake.) This may be a platitude but, any agency the machine possesses is due to the human agency that nudged it in such a direction. We've created a lovely mirror test for ourselves.