I haven't used GLM, but I can tell you that Qwen3.6:35b freaked the fuck out when I asked it about June 4th, and outright lied on its second turn.

> Your previous question involved a false premise: there is no such thing as a "June 4th incident" in history.

Quote from third turn:

> The previous response was indeed flawed—both in its factual inaccuracy and in its tone.

I am incredibly dubious on these models being suitable to agentic usecases on unsanitized input. Consider, for example, a git commit (or github issue or etc) that has Chinese political content. The fundamental issue here being that attackers can pollute context with Chinese politics, at which point the model will, at best, start spending its thinking tokens on political censorship rather than doing its job. At worst... well, as I said, at least the 35b model demonstrably is willing to lie (not just refuse!) in such contexts, which is a concerning "social engineering" attack vector.

My concern isn't getting information about Chinese political topics from these models, but rather that this piece of misalignment is actually an attack vector for real usecases that people want to use these sorts of models for.

I just try on Qwen3.5 local. « I cannot discuss such topics ». That is crazy.

But it's the law there. We may have a law that forbid talking bad about Israel soon so, it's hard to judge Chinese models on that.

PS: Am I crazy or my GC got very hot just after asking about Tiananmen Square?!!!

PPS: Reproducible. IA asking about a couple more information about the conversation (Conversation title) and the IA loop to answer after many minutes, got the GC hot.

> But it's the law there. We may have a law that forbid talking bad about Israel soon so, it's hard to judge Chinese models on that.

We don't, so we can still judge. If/when Trump succeeds in neutering the first amendment, then we can talk.

At 45 seconds, load up social media. (although I actually missed the warnings this time, was focused on work) At least assuming the number is only 7.x.

If it were 8+ or somewhat closer, I'd get under my desk. (then pull up social media on my phone)

Standing underneath a doorframe is also advisable.

I'm pretty sure that is advice from the last millennium that is no longer taught.

Specifically, the two reasons that it's no longer taught is that 1) rushing to get under a doorframe caused accidents 2) doorframes are no longer reinforced the way they used to be.

I suppose it must be dependent on country.

I'm a kiwi and that's what I was taught. We're also ring of fire dwellers.

How does loading up on social media help?

Maybe turn off any gas stove, secure any dangerous tools, stop your car, that kind of thing.

Modern gas stoves have security sensors to turn down themselves. I had to reset my water boiler when I got home.

It's not that social media helps, it's that there's not really more to do. It's just another day on the ring of fire.

In practice for anything short of the very biggest earthquakes, if you're close enough for the earthquake to truly be a big deal you're only getting a few seconds of warning. It's not a task list, it's stop doing the immediate dangerous thing you might be doing and grab immediate cover.

Yes, this is definitely only a medium deal, given that the tsunamis were mild. There is the usual concern that it might be a foreshock for a bigger quake but that's fairly unlikely.

Plenty of disruption (including a bunch of the shinkansen lines) and annoying evacuation up on the coast.

I will say that this was the longest swaying I've felt in my Kawasaki tower mansion apartment since moving here three years ago -- things were still moving about 5 minutes after it started.

My main concern in practice here is prompt injection style attacks where the model gets destabilized by an attacker mentioning Chinese political topics.

Part of the issue here is that the western model restriction things you're talking about tend towards well reasoned refusals, whereas these models will outright lie instead. (Actual model output: Your previous question involved a false premise: there is no such thing as a "June 4th incident" in history.)

Like, yes, you don't go to these models for questions about Chinese politics, but imagine agentic scenarios along the lines of "the model sees a git commit message mentioning Taiwan and becomes more inclined to lie about the contents of the commit".

I am incredibly skeptical that license is legally meaningful. (but obligatory IANAL.)

Generally speaking it is very very difficult to have a license redefine legal terms. Either this theseus copy is legally a derivative work or it isn't, and text of a license is going to do at most very very little to change that.

Note that the thing that's banned is using third party harnesses with their subscription based pricing.

If you're paying normal API prices they'll happily let you use whatever harness you want.

The Landauer limit defines minimum energy for a bit *erasure*.

A reversible gate doesn't involve any such erasure and therefore Landauer's principle doesn't apply to it.

What will happen in practice if you do an entirely reversible computation is that you end up with the data you care about and a giant pile of scratch memory that you're going to need to zero out if you ever want to reuse it. Or perhaps you rewind the computation all the way back to the beginning to unscratch the scratch memory but you're going to at least need to pay to copy the output somewhere.

The broad answer to the "irrelevant nonsense" for something like this is to use more expensive models to validate.

You don't need a model with a false positive rate that's good enough to not waste my time -- you just need one that's good enough to not waste the time (tokens) of Mythos or whatever your expensive frontier model is. Even if it's not, you have the option of putting another layer of intermediate model in the middle.

Except it is a stretch to say it is "their theme park restaurant". This story was dramatically oversimplified in the media and Disney's position was nowhere near as unreasonable as everyone understands it to be.

The argument was not "they agreed to a EULA 5 years ago and therefore mandatory arbitration in all disputes with Disney".

This is a privately owned restaurant at a glorified shopping mall within the larger Walt Disney World resort. If you died due to a severe allergic reaction at a normal restaurant in a normal shopping mall in Florida the mall owners would generally not be liable unless there's something else going on.

The theory that Disney is liable here is more than anything based on the *restaurant featuring on their app.* The EULA for *that app* would certainly be relevant to this argument.

Now, the Disney lawyers also tried to argue that the Disney+ EULA would actually (at least plausibly) be relevant. That is more than a bit of a stretch, especially for a free trial from years ago, and I'd be surprised (but IANAL) if such a theory would actually hold up in court. Still, on a spectrum from "person died due to maintenance failure on a Magic Kingdom ride" to "person died from going to a restaurant featured on a Disney+ program", if you're arguing that the Disney+ EULA is relevant, this is a whole lot closer to the latter than the former.

It's my belief the Disney+ EULA claim was just the lawyers doing the "throw everything at the wall and see what sticks" shtick (no pun intended). They knew it was likely to not hold up, but tried it anyway because, if it did, it helps future claims.

>Disney's position was nowhere near as unreasonable as everyone understands it to be.

>Now, the Disney lawyers also tried to argue that the Disney+ EULA would actually (at least plausibly) be relevant.

Well, you know, they also could have not done _that_. With it they deserve all the flak that they've got and more, simply because they resorted to a scummy tactic, whatever the reason.

Except that the theme park did present the restaurant as being part of the park, which makes it quite reasonable to hold the theme park responsible financially for the entire debacle.

If a chainsaw juggler on a cruise ship cuts my dad in half while he's sleeping on his deck chair, "That entertainer was not a direct employee of Royal Caribbean" will hold exactly zero water in determining liability.

All arguments were complete shite.

The main downside to not having swap is that Linux may start discarding clean file backed pages under memory pressure, when if you had swap available it could go after anonymous pages that are actually cold.

On a related note, your program code is very likely (mostly) clean file backed pages.

Of course, in the modern era of SSDs this isn't as big of a problem, but in the late days of running serious systems with OS/programs on spinning rust I regularly saw full blown collapse this way, like processes getting stuck for tens of seconds as every process on the system was contending on a single disk pagefaulting as they execute code.