vintermann

I have wondered how far you can get without an engine and with computer stabilization. A plastic bag flies in the wind, and sailships can sail upwind to some degree, so surely someone is experimenting with solely wind powered drones?

I interviewed for a company which used time series modelling to predict engine failure on ships. Though they didn't call it that, they called it just "AI" and weren't familiar with the time series literature at all. (Neither was I, but I did at least know that the time series literature was a thing.) They just used a bunch of sensors attached to the engine.

That seems like a scenario where there wouldn't be external sources of information capable of improving your predictions. It also seemed to me like a scenario where good old fashioned statistical methods would outperform LSTMs, which was what they used.

(I didn't get the job - I was offered it, but they rescinded their offer at the last moment, supposedly because their investors had insisted they should hire a web developer instead.)

That claim doesn't seem to be sourced. It sounds a little dubious to me.

Heh, Prism Launcher, the author's example, is the main app I use flatpak for currently. And it works OK, except that I can get some very strange graphical bugs if the NVidia driver updates and I forget to update the flatpak Nvidia runtime.

That's wild. Yeah, a quick look into the background of this author leaves little doubt about what their career is.

I'd never heard of the BabyVision benchmark they do so well on. Pretty interesting stuff:

https://arxiv.org/abs/2601.06521

... That would explain a lot. I'm in Norway, and I had never heard of these biscuits until recently. Now they're absolutely everywhere. It's maybe not surprising that the US fast food chains push them, but the Swedish chains do too? And they put it in chocolate bars?

Like most countries I'd wager, we have plenty of own local biscuit brands you've never heard of. Biscoff does not stand out from them as fundamentally different in either texture or taste. I just couldn't wrap my mind around why this bland brand was suddenly pushed so heavily everywhere.

But if food executives think it's GLP resistant (whether it is or not) then it makes a lot of sense. Jesus. People should worry less about "hyperprocessed food" and more about things like this (alternatively, this sort of thing is one of the hyperprocessed foods people should actually worry about).

But why are we all going to standardize on Meta's model, and how is it sticky? What does it have on the big Chinese models?

Durkheim's classification works as a description, but doesn't seem very good to predict which groups will have high suicide rates. Suicide follows some very odd patterns. For instance

* Black Americans have low suicide rates, lower than white Americans, who in turn have lower than indigenous Americans.

* The children of actors and celebrities have high suicide rates.

* Men have higher suicide rates than women.

* Wealthier countries seem to generally have higher suicide rates than poorer countries, but inside a country the poor usually have higher suicide rate than the rich.

* People with schizophrenia have sky-high suicide rates

The only common thread I can think of is a sense of agency - which means thinking you have many options - with hopelessness. When you can do a lot (or at least you think you can, or society thinks you can), but none of it helps with the things that matter to you.

* Victims of racism may paradoxically be somewhat shielded from suicidal despair from low expectations. But when racism comes with heavy romanticizing (e.g. "noble savage", "ancient proud culture"), expectations are not low. Yet, it's all too understandable why someone would want internalize these romantic, often openly envied identities, even when they're completely at odds with their material situation. You can't live up to it yet you feel like it's all you have.

* Children of e.g. wealthy actors are materially very comfortable and safe, and clearly have lots of agency and choice - but what do they want, when they don't need to strive for wealth, any social acclaim they earn may be tainted by the suspicion that they didn't earn it fairly, and they in any case are unlikely to surpass their parents?

* One typical schizophrenia symptom is sense of agency disturbances. They often feel like they are in control of things they aren't, or don't have control of things which they do. The former, combined with seeing bad things happen (for instance on TV) can really devastate them.

Right, and I definitively don't want to guess at what I might want. A "round trip to the user" is much better than getting bogged down implementing something I didn't ask for and don't want.

How many of these features would you expect to also see in children's drawings? I have a hunch that the sun in the upper left corner for instance is pretty overrepresented.

Okay. Can you think of some groups which have a low sense of agency/few options, yet have a high suicide rate because of their lack of purpose and identity?

I'm looking for evidence against this theory, so I'm very interested if you can find it. But I think it's going to be hard, because they're going to be correlated: if you have a weak sense of self and purpose, then very likely your options will all seem bad too. And even if you have a strong sense of who you are and what your purpose is, failing to live up to the identity society and yourself assign to you (because all the options you have fail at living up to it) then you could well call that "having your identity taken away from your".

I'm sure some sociologists have written this already, but it seems to me that the people who commit suicide aren't those who are miserable and have few options. It's those who have lots and lots of options, but all options lead to unacceptably bad outcomes.

Microcredit in the third world is somewhat notorious for causing a rise in suicide rates. That fits the pattern: there's a huge number of things you can do to get money to pay back the loan, but if it gets out of hand, none of them get you remotely enough money to keep from sinking deeper into debt. Access to capital gives you many options - but it also makes it your fault when you didn't find the option which would lead you out of miserable poverty (which may not even exist!)

What any psychologist will probably tell you (at least, what all of them told me) is that clinical depression makes you bad at seeing options. Could it be that's in some sense a defense mechanism, trying to stop you from putting the really bad options on the table?

A shame. It would probably have been a good thing for the world if we knew exactly what Open AI executives were talking about.

More clear title, given how people talk about AI:

"Capabilities can be hidden in AI model weights by encryption".

It's not about AI agents being able to hide what they can do.

Farmers did well compared to hunter gatherers, but they were in turn devastated by pastoralists - who, in Europe at least, had a lot of hunter gatherer ancestry.

This is just one study, which the website notes goes against consensus, and we have a very sparse view into ancient DNA. Just from knowing Norwegian geography, I'm rather skeptical of full-genocide theories.

An older model that didn't yet have obfuscated thoughts? I'm skeptical. I think the non-obfuscated CoT traces I've seen all look similar.

That's right, what matters to me is what matters to me, and the something you've measured I don't know - but that's not a point in your favor.

The worst sin a model can commit in my opinion, is to give an excellent dazzling response to a slightly different assignment than the one you gave it. DeepSeek seems really good at NOT doing this.

But if you ask the model what it expects to be asked, of course you won't have that problem. It could of course be that DS commits this sin, but just happens to expect the tasks I give it.

But I rather think that it's Claude which is good at expecting your tasks - because I have seen all your "high standards" models commit this sin.

Two theories:

* It was off-the-charts brazen, and amazing that so many prestigious people got behind it, and that it took so long for anyone blow the whistle/notice that the impossible product didn't in fact exist.

* Sexism.

If you want me to believe #2, I would like to hear what you think are the more comparable cases which don't get bought up as much as this one.

I have a theory! They saw all the other security state people involved, and figured this was something it would be a good career move to support without knowing exactly why.

"Don't obey in advance", Tim Snyder said. But obeying in advance is called showing initiative and is a great career move. In secretive industries it can be hard to know what to obey in advance, but they don't let that stop them!

In that case generative AI could be defended as a privacy preserving measure, I guess.

What they pretended to do more like it.

Mistakenly and autonomously of course.

It's better to look at the results than the CoT. As far as I know, the CoT is censored for US frontier models - it certainly was for Gemini last time I tried. When you get a condensed summary of the CoT omitting all the false leads, incoherent digressions and backtracking, of course it's going to look smarter.

> I've benchmarked, rigorously, deepseek-v4-flash for programming and personal use

You've measured something, but I'm not convinced you've measured what matters, because that's a lot harder than people give it credit for.

I remember seeing the QuakeC line of code that halved self-damage from rockets. It enabled rocket jumping, but it also made the rocket launcher a much, much better close quarters deathmatch weapon than in Doom, so I thought it was a bit lame.

All these products, western or Chinese, are built on a hell of a lot of running rough-shod over licensing or IP laws in general.

I'm writing a program I personally need, but I would be happy if there existed something like it already, if someone else vibecoded a better version of it than mine, or if DS got better at vibing this kind of thing.

Pretty! I checked OEIS, and the "interesting" tilings appear to be for numbers k where k < floor(sqrt(k)) * ceiling(sqrt(k)) (A189151).

Probably obvious for people who try to find these tilings, but I thought it was neat.

A Swedish proverb says, "a fool may ask more than ten wise may answer". This fool is reporting for duty! I'm glad I may have something to contribute after all (and I'm only halfway joking)

What about cheap solar panels and electric cars, is that economic warfare too?

To say that US model providers have a close relationship with their government would be putting it mildly.

Just go find a list of vaccines? I followed a tracked of vaccines which were under development during the pandemic, that tracker appears to be down, but even Wikipedia claims at least 50 got far enough to gain approval in at least one country.