@sapinker There has been a lot of debate around these issues, and there are very basic rebuttals to the claims you raised. Could you address those maybe? Else I'd suggest not commenting on this topic in public. This article is strong evidence you haven't actually thought very much about the topic. Like notably === 1. Clearly some systems are more generally capable than others, and this lends them power. You'll be hard pressed to find a task gpt 4 can do but gpt 5 can't. And e.g. the huggingface incident displays agents having more power over the world, in a way earlier models wouldn't. There are theoretical arguments around no free lunch theorems, but they clearly are not the kind of theorems you can straightforwardly apply to our world, because applied fairly, they'd not predict humans taking over the world. 2. Again, just false. Motivation and intelligence are importantly different. And x-risk people know this. There is a reason they talk so much about orthogonality and alignment. The argument is that the current method of creating AIs does produce agents with (unintended) goals, which is clearly true. (empirically! although was obvious in advance.) 3. Again, is just a straightforward misreading and has been addressed ad nauseam. Paperclips are used in the thought-experiment because it is pedagogically simple. But you could recreate the thought experiment with an AI that in addition to paperclips, also wants to talk people into buying legos, computing that largest primes, writing as many unique fugues as possible, and which has a complicated scheme for weighing these separate goals against each other. Such an agent would still disempower humanity. 4. Nobody assumed this. People assumed (argued that) a smart enough AI would be able to take over without being handed anything. But actually, it turned out that people are handing the AIs control over everything anyways. So even if doomers were wrong on point (4), they'd still be right! === I'm not really asking you for object level responses to these points. Just making the meta-level point that, these are very obvious and elementary responses, which I've had cached in my head for many years, which I'm confident a large share of those worried about extinction would raise in response too. And therefore not addressing them just makes it seem like you've not engaged with the arguments at all, and therefore that I should not give your opinion much weight, and furthermore that hearing you comment on it publicly, should update me negatively on your epistemic rigor (and other things).