Just because someone does a bad thing with a tool, then turns around and says he didn't do it... well that's not a good reason to believe them. In the real world, we call these people criminals and hunt them down, we don't say "oh, the tool messed up, could have happened to anybody."

Here's your answer.

No, this one.

That's not what Anthropic promised their shareholders. If that was all Claude is supposed to be, nobody would be giving them that much money in the first place.

The problem with all these arguments is that they beg the question: They assume that AI is some kind of sentient technology with personal and unpredictable goals that are inevitably in opposition to humanity's goals. They then argue that humanity is controllable because it cannot fight a super intelligent sentient technology with personal and unpredictable goals. It's a classic logical fallacy.

There can be no co-operation without some measure of trust. It's a collective balancing act which is not purely technical. AI mimicry enables new kinds of social engineering attacks that have never been possible before. The onus should not be purely on the software architecture, IMHO.

Agreed! I used to RUN a 4KB BASIC interpreter in my T-16 back home!

I'm nit optomistic that thonking thos fast is a giid odea!

My Emacs only needs 8Mb! (but it's constantly swapping)

Those issues are not new. What *is* new is that AI mimicry has lowered the bar for attacks substantially.

Just creepy crawlies looking for their 5 minutes in the spotlight while the audience is ooh-ing and aah-ing the destruction and opening their wallets to buy trinkets.

Them's there alternative facts meester! I gots plenty! It's see-ent-tuffic, y'see!

obligatory gravity well explainer

There's no overarching EU law. There are EU regulations and directives, and the member states (who each have their own state laws) must fold those directives into their own state laws in a way that fits. The regulations tend to be very targeted.

So in a manner of speaking, it's all state laws, no "federal" law, just local interpretations of "federal" directives and some common standards. And contracts in each state have to follow state law. If someone objects that a "federal" directive is broken, then they can sue the state in an EU court, etc.

A big difference with America is that a legal precedent in one state doesn't mean anything in another state. It often doesn't mean anything in the same state either. The judges interpret the legal texts, but do not create new case law. You can't refer back to some judge such-and-such who said something was ok in a similar lawsuit, so therefore it must be ok going forward in all future lawsuits.

That's a meaningless statement. The politicians frame the engagement, but that has always been the case in all wars everywhere. Read Clausewitz. There are parameters and objectives. If a general can't deliver within these parameters, it means he's not good enough.

Same if your boss is telling you to build some accounting software, and you complain that it's impossible and he should let you build a flight simulator instead.