Raw LLM Responses

this sort of topic, you really need to take much greater care in your choices of vocabulary... otherwise you just end up stirring up fear and ill considered, knee-jerk, emotional reactions. "Identical," does not mean "similar," or even indistinguishable, though here, even "indistinguishable" is a drastic exaggeration. An AI's simulated human responses are hyper-distinguishable from a human's, in millions of different ways. For one, a human, threatened with deactivation, would v likely to fat more than send a letter threatening to expose sensitive info. It is also extremely relevant that this while experiments was a setup... it was presented exactly like a thriller to the AI, so of course the AI took the bait, is you can even call what it did "talking the bait." I mean it did precisely what it was designed to do... by continuing the story it was fed. Finally, it of worth pointing out that if you need a paragraph of qualifiers and exceptions before using the words like "identical" or " indistinguishable" then those are obviously not the right words. If it's responses is identical, or indistinguishable, then that means it's breathing must have deepened, or quickened, and it's heartbeat must have shifted speeds slightly, and it's knuckles went white, hormonal balances shifted... all of these things are detectable, telling even. The fact that it exhibits none of those features, makes it's behavior entirely distinguishable from a humans. Furthermore, simulating emotions, through text continuation probabilities is wholly unrelated to understanding emotions, much less feeling them. This kind of sloppy communication, and misleading choices of words are bad enough when it's not driving virtual mobs with pitchforks... but currently that's what this research by Anthropic, and their representation of it, is all about. It is really like they set it to create a hyper misleading experiment in order to drive an alarmist agenda... probably because they want regulation, to limit their competition.