Can you mathematically prove it’s impossible?

Being able to prove a positive doesn’t imply being able to prove a negative.

A year or two they couldn’t even fool the average viewer though. To prove a video is real you’d also have to prove there’s no model out there in the world (including in private testing) that could create a good enough fake.

He’d be a moron to do that tbh. Right now the US is massively profiting off the high energy prices. Not its people but its oil companies at least.

Huh I thought they didn’t want foreigners tho

How would that solve tourism? There should ideally be no tourism under socialism, it’s a waste of resources.

They only dislike Muslims who aren’t billionaires.

Protecting New Yorkers of various backgrounds is real work for the mayor of NYC though?

And judging by the hate crime statistics he brought up, people also need to realize the reverse: that Israel doesn’t actually represent all Jews worldwide.

Which is really the same thing tbh.

Long as there’s demand for bonds, they won’t really ever pay off the debt. Matter of fact, having debt is like the default for governments because it’s cheap money if used wisely (so not whatever the US is doing). Right now the average interest rate on their debt is still lower than at the start of the century. Trump needs to keep hurting the country to make it high enough to matter, but he’s got over 2 years to do it so I’ve got full confidence in him.

Neither is any LLM, nor any mainstream agent I know of. You CAN use them that way IF the model doesn’t trip a safety guard, but it’s gonna take a bunch of prompt engineering to get most of them to do anything illegal. I could barely get Deepseek to reverse-engineer offline software by asking for it.

Anthropic famously has a separate model for cybersecurity analysis because they won’t let you do shit on even your own infra with the publicly available Fable. You’ll have a hard time even analyzing your own source code for security vulnerabilities with Fable for that reason, though Opus might work.

I seriously doubt any of these agents are actually hacking websites from a “I need XYZ information” prompt. Claude triggered the ol’ “need to downgrade to Opus for safety” on me when I wasn’t even asking it to investigate anything security related, just help track down a few bugs.

The whole “our agent went rogue” thing is almost certainly marketing. They WANT you to think it’s dangerous, they’ve been saying it since like GPT 2 which couldn’t produce a paragraph of coherent text. They WANT to be regulated because that makes it harder for smaller players to compete. Because right now z.ai’s GLM 5.3 Flash is about as good as Claude was a few months ago and it costs pennies in comparison (possibly subsidised by China - who knows).

“I’m going to continue as long as I am chairman to resist this anthropomorphizing of these tools,” Ferguson said at the Reuters Momentum AI Austin event. “If someone tells a tool to do something, and the tool does it, I don’t think we would say, ‘Oh, what do we do about the tool?’”

To me that says it’s the user

Is the manufacturer of a knife liable for stabbings?

So if I use curl to hack a government agency, is the developer of curl liable too, or is it only LLM companies that are at fault if end users decide to do something illegal using their tools?

Read again.

“I’m going to continue as long as I am chairman to resist this anthropomorphizing of these tools,” Ferguson said at the Reuters Momentum AI Austin event. “If someone tells a tool to do something, and the tool does it, I don’t think we would say, ‘Oh, what do we do about the tool?’”

AI companies aren’t going to be responsible for shit, only the people using AI to hack things.

As I understand this would mean if you run OpenAI Codex and instruct it to do something and end up hacking someone, you are at fault. Not OpenAI. So be careful what you 100% agree with.

Those are the models. An agent is a harness around the model.

So if this goes through I could use opencode with GPT and… Which one is at fault, Opencode or OpenAI? What if I also modify Opencode’s system prompt as a joke to make it as unhelpful as possible and forget I did it, am I at fault or is it still Altman who goes to jail if I now let it delete my production database?

As I understand it, it’s the person running it, so if you let an agent fuck something up, YOU are at fault, not the developer of the model or the tool. You’re the one who gave the task. You have the ability to block harmful commands if you don’t go yolo mode and actually read what it’s doing.

The entire world was doing lockdowns though. Not just the US.

You were already told that the number that wants the war to end doesn’t all want to surrender 🤡 there’s a difference between wanting the war over and wanting to live (and die) under Russian rule.