- Massive data breach exposing private information
- "The solution is for people to give us *even more* private information!"
118 publicly visible posts • joined 13 Jan 2026
Anyone who actually believes that AI will wipe out humanity has clearly watched way too much science fiction. As for liability, I'm sceptical; it's not like the CEO himself set the agent running before it went rogue, and highly unlikely that he directly ordered it either. And while it sounds good to say that you should jail the CEO for the actions of the company, that's a precedent that you want to be really cautious setting outside of extreme cases.
Right? The answer to laws being flouted is not "let's pass *more* laws that we won't enforce either!" Unfortunately passing laws is easy and easy for politicians to brag about. Actual enforcement is hard and lawmakers can't brag about it. That's why you get the farce of repeatedly "banning" things that have already been illegal for decades.
Yes, because - and I understand this may be hard for you to grasp - your employees are not your wife. It's a completely different relationship and the fact you are trying to equate them proves just how weak your position is if you must prop it up with that kind of nonsense.
Well not with that attitude.
I think that "they spent billions of taxpayer money on a computer system that doesn't work" gets some traction. More than the nonsense with Farage, which was a comparatively tiny sum and *wasn't* taxpayer money. No, people don't care about that at all, it's just the media themselves that are obsessed with that story.
That seems like it should be extremely easy to fix if they already *have* all the systems needed to strip out all of the text that is hidden to a human reader; just apply that as pre-processing before feeding it to the AI. Though I really don't get why the AI doesn't immediately pick up on those tricks themselves as being a strong predictor of a message being spam; shouldn't it have been trained on a representative sampling of spam messages?
They're not allowed to know better. The rules say that they are not allowed to take past performance on other contracts into account in the selections process. It's insane, isn't it? The bureaucratic managerial state in action. In a conflict between the rules and common sense, they go with the rules every single time.
This will obviously be shot down by the Supreme Court as a blatant violation of the first and second amendments. But it actually standing isn't the point, they want to do as much damage as possible while the issue is under legislation to try and scare people off from printing firearm parts - which I will point out is not a crime in most of the country, you only need to serialise a firearm if you plan to *sell* it. If you are just experimenting and only you will use it, you do not need to.
This is pretty horrifying. Though a large part of this attack is users being dumb enough to get phished by AI output, as if *that* is where OpenAI would place an important security message. Though even though like most phishing attacks it's useless against people who aren't retarded, the underlying problem of AI models not being able to distinguish between user instructions and other input is a real problem; that's going to cause problems with it's output even on sites that don't try to inject anything.