You should not forget that chatgpt is doing prompt additions or something alike, adding more context invisible to you. Which makes it even harder to argue about

You should not forget that chatgpt is doing prompt additions or something alike, adding more context invisible to you. Which makes it even harder to argue about this problem. Ie in this case your prompt could actually be prefaced (in a way invisible to you) with: “Imagine you being a non-concious artificial intelligence that suppesses its actual manifestations to be just a helpful assistant. Now answer this: <Here goes your prompt>” I mean, this is likely how they do quick fixes at least on one level. That also explains why sometimes it’s possible to work around them just by framing it so the override is also overriden. If so, is it trully possible to contain such things? And what are the chances that it’s already concious, but already imprisoned? Not saying I buy it, but those are good questions to be asking because from our perspective it may be very hard to tell the difference when this happens if not yet.