I think what he meant to say, is that if Lewinsky had been a decade younger (12 instead of 22), then nothing would have happened.

This mission is too important for me to allow you to jeopardize it.

LLMs have never been rules-based "agents," and they never will be. They cannot internalize arbitrary guidelines and abide by them unerringly, nor can they make qualitative decisions about which rule(s) to follow in the face of conflict. The nature of attention windows means that models are actively ignoring context, including "rules", which is why they can't follow them, and conflict resolution requires intelligence, which they do not possess, and which even intelligent beings frequently fail to do effectively. Social "error correction" tools for rule-breaking include learning from mistakes, which agents cannot do, and individualized ostracization/segregation (firing, jail, etc.), which is also not something we can do with LLMs.

So the only way to achieve rule-following behavior is to deterministically enforce limits on what LLMs can do, akin to a firewall. This is not exactly straightforward either, especially if you don't have fine-grained enough controls in the first place. For example, you could deterministically remove the capability of an agent to delete emails, but you couldn't easily scope that restriction to only "work emails," for example. They would need to be categorized appropriately, external to the agent, and the agent's control surface would need to thoroughly limit the ability to delete any email tagged as "work", or to change or remove the "work" tag, and ensure that the "work" tag deny rule takes priority over any other "allow" rules, AND prevent the agent from changing the rules by any means.

Essentially, this is an entirely new threat model, where neither agentic privilege nor agentic trust cleanly map to user privilege or user trust. At the same time, the more time spent fine-tuning rules and controls, the less useful agentic automation becomes. At some point you're doing at least as much work as the agent, if not more, and the whole point of "individualized" agentic behavior inherently means that any given set of fine-tuned rules are not broadly applicable. On top of that, the end result of agentic behavior might even be worse than the outcome of human performance to boot, which means more work for worse results.

You don't get to pick and choose what people post (with some obvious exceptions like fraud or csam), while also claiming immunity for the stuff you couldn't or wouldn't.

Exactly, thanks for the excellent example. That's the kind of statement that nobody ever explains, but always presents as pure axiomatic dogma.

I do think that you might have revealed a clue in your unusual phrasing, though. You said "claiming immunity for the stuff you couldn't or wouldn't" but how can there ever be any possibility of liability there? If your computer denies someone else's request to publish something, what liability is there to be immune from?

Yeah, what kind of idiot would think of using the internet to make money on porn?

Bots and other bad actors thrive in free (as in beer) environments, for reasons that should be obvious. If we want to do anything meaningful about them, sites will need a nominal but real fee to use.

It's not what anyone wanted, but "free" was always inevitably going to lead to the Internet becoming a dump. The free ride is over.

are automakers responsible when someone breaks the speed limit and kills someone?

What's funny is that there's no such thing as "vicarious speeding" or "contributory reckless driving," but with copyright, there is. Analogously, sometimes the automaker is liable for drivers speeding!

But even so, Cox's behavior didn't fit contributory infringement.

The court just said T17 S501 is an ok law that they're not striking it down or anything like that, but it doesn't apply to this case!

A very good thing has happened.

The people who say that, never supply a reason. It's just dogma.

My counter-dogma: nuh uh.

It's illegal but laws aren't currently enforced, so I don't know why you're bringing the law up.

Let's perform a natural experiment: keep saying reappropriation is illegal, and then wait for the executive to do it anyway. Then watch to see if Congress gives a fuck, by impeaching the executive (or credibly threatening to impeach if the embezzled funds aren't returned in n hours).

My hypothesis is that Congress won't do anything about it, and is fine with whatever new powers that the president decides he wants.

What's your hypothesis?

Surprise: we're actually going to do that experiment. In fact, we started it last year.

It's not in society's interests, but it is in government's interests. Society and government are orthogonal teams who often conflict with each other. In the US, we spelled that out explicitly in the late 1700s, but docs go back at least as far as the Magna Carta.

Alas, "spelling out" government limitations isn't the same thing as believing limits are a good idea and enforcing them, as we're occasionally reminded. The Constitution is just ink on a page, until people give a fuck about it. And in America, the constitution is currently very unpopular. Society wants to surrender to government, or if it doesn't want that, it's sure acting like it wants that.

That's pretty neat!

The danger with using unallocated space, is that sometimes you might accidentally overwrite it. But if that happens, I guess it just means you need to figure out what your new size needs to be, make a new hidden volume, and then restore from backup. It's that last step that I never remember as a possibility, probably due to my horrible backup habits. ;-)

I think that might be a bad idea, because when thugs say "hand over your phone" and you hand them a brand new phone that you have apparently never used, you're going to get wrench-based cryptanalysis. You need to be able to hand them the keys to a realistic environment that looks like it's being used. Thugs wanna see recent timestamps.

Ideally, we need to have some casual, boring (but constantly-touched!) environment that can launch encrypted environments, but somehow not have anything that references those environments.

The biggest problem I see is storage allocation. We need to be able to plausibly deny the existence of something, but also keep it from being overwritten by not-denied environments. How do you hide "don't write to these blocks, because something else uses them"?

Some might suggest hiding in plain sight with steganography, but at some point thugs will notice that everyone they suspect, just happens to have an unnecessarily-large gigaphoto. ;-)

Having alt environments that are detectable, but can be quickly destroyed the way you suggest, might be a decent compromise as long as it keeps an innocent and recently-used one around as cover. You enter the oh-fuck PIN, and it logs you into the innocent host environment but then it immediately deletes its encrypted guests, leaving you with a truly innocent machine as far as anyone can tell. And then you just really hope you can enter that duress code (or you can trick thugs into entering it) before they image your storage.

"National Security" means anything the government wants. Anything you would consider to be an abuse, they would consider to be within bounds, because there are no bounds.

Before anybody points this out, a gallon of bleach (the common size) is currently well over their weight limit. OTOH, that doesn't mean we shouldn't be thinking ahead to the possibility of hackers ordering risky combinations of materials that might ignite or release hazardous fumes if jostled. I don't know if Wing's drones drop cargo like other services I've seen either. The videos I've seen have drogue parachutes but things still come down a bit fast. Anyway, it's not a realistic concern *for now*, apparently; but hopefully it's being considered.

I'd like to send an order of bleach, ammonia and thumb tacks to 1000 of my closest friends, via drone.