atq2119

A lot of the public successes with agents is really LLM-driven local search against an objective function that is evaluated in more traditional ways. This one seems to fit the pattern.

I'm okay with that if it helped prevent Nazi Germany from building an atom bomb...

This is the kind of thing that really needs a primary source instead of ChatGPT.

You're not crazy, economists really have a weird an IMO misleading way of drawing their diagrams.

I agree with you that price should be on the X axis, and drawing the two supply lines in the same diagram is at least somewhat problematic.

What economists posit[0] is that at any point in time, there are demand and supply curves. They answer the question of who is willing to sell or buy how much given a price? (Quantity is the dependent, so should be the Y axis!)

And they argue that the microarchitecture of the particular market causes price and quantity to converge to where these lines intersect.

And then factors external to the market can change the supply/demand curves. The February diagram looks different from the April diagram. They are conceptually separate diagrams. Combining them into a single diagram in a coherent way would lead to something 3D, which is hard to draw and think about, so economists have the convention of drawing it all in a single diagram anyway.

None of this is correct, by the way, but it's sometimes a useful model.

[0] Outside of literal markets with order books, supply and demand curves don't really exist. And in those markets, their dynamics are different.

In order for the ASIC to achieve 1% of the price, there'd have to be 99% overhead in GPU implementations which for some reason you'd have to be able to eliminate in ASICs but not in GPUs. That seems rather implausible.

FWIW, what you're describing is a simple demand curve, not Jevons paradox.

The "paradox" is when an increase in efficiency which would decrease the use of a resource all else equal, instead indirectly causes more use.

That's a feel good response, but there are also individuals who post a bit too much, but still fairly interesting stuff.

A middle ground would be to have a "reduced follow" mode where you still have a chronological timeline, except that at most N posts per day are shown from "reduced followees". The filtering can still be based on a popularity signal.

Nothing in your link is evidence of a problem.

It also doesn't explain the regional correlations. If immigrants are objectively a problem, why is anti-immigration sentiment highest in regions with the least immigration?

Make it make sense...

> the immigration issue that clearly is forefront on the minds of such AFD voters,

Let's be real: this "issue" is a strawman. It doesn't really exist, especially in that part of Germany. Just like pretty much everywhere else, the German anti-immigrant far right is strongest in the areas with the fewest immigrants.

This is not to say that the wave of immigration that is now more than 10 years past didn't have its challenges, and they were downplayed by some.

But at its core, the immigration scare is just the classic playbook of offering simple but wrong explanations to distract from the class warfare that's being waged by the wealthy.

Ideally, we'd turn down the temperature and have patient conversations, but the media environment was made very counterproductive over the last few decades.

It feels like this progression of increasingly drastic measures to circumvent the protections of a computer system ought to be enough to establish criminal intent and get some of the people running those crawlers into prison.

This is ultimately about power.

Democracy is built on the idea that people are equal and have equal rights. This is only sustainable if the power differential between people isn't too large.

Corporations consist of people, yes, but the power to control where the efforts of these people are directed is concentrated on very few people, giving those people a large amount of power -- so large that without any checks on their influence on government, it undermines democracy.

It's the same argument for why super-wealthy individuals are so problematic. They're literally destroying the fabric of democracy through the power imbalance afforded by their wealth.

Assuming that summary is accurate: it specifically talks about businesses, so the requirements presumably don't apply to a personal website, which is what the post you were responding to was about.

I want to agree with the sentiment. Then again, OpenMP has in fact used the term "team" in the this kind of sense for a long time.

I suspect it's a similar phenomenon as people capitalizing visa when the topic is international travel / migration instead of payment networks.

As a non-GCC compiler developer myself, there's a spectrum of AI use and the words "AI generated code" can mean different things to different people. At the extreme ends, nobody I know well objects to AI auto-completion in your editor, but neither does anybody want to let AI open pull requests autonomously.

Having a coding agent iterate on a well-defined narrow task in the background while I'm in a meeting or working on something else, and then doing a thorough, thoughtful, and active local code review (meaning, having an editor and the diff open side-by-side and liberally making edits to clean up the code) before pushing anything out for others to see feels like a reasonable point in the space right now. Nothing is perfect, but this style of active code review removes basically all AI smells in practice. The edit-build-test loop is fairly long (i.e., longer than a few seconds), so having AI babysit this loop for the initial development of a task is helpful.

The sad fact of life is that too many developers are used to pushing PRs with unprincipled commits, and relying on squash merges.

The approach that GitHub chose can show those people almost immediate benefits, which is surely the better path towards adoption than trying to reeducate everyone to adopt a development flow based on clean rebases.

And for those of us who do prefer clean rebases, the thing they built is still useful.

A bunch of details aside, the major conceptual thing that the email based flow has which is still missing in GitHub's data model is the ability to have a discussion on the stack as a whole.

Do you mean on Linux in a Wine/Proton context? That's different, so yes, I did mean DX12 driver.

Again, not impossible, just picking your battles.

In order to have a useful Windows desktop, you need a DX12 driver.

So, in order to write their own KMD, they'd either have to also write their own DX12 driver, which they probably have no interest in, or reverse engineer the existing AMD KMD interface to the point where they can reimplement it basically completely so that AMD's proprietary DX12 driver runs on it.

I doubt they're saying it's impossible, it's more of a pick your battles situation.

I sort of see the point about undoing the changes to the clipboard, although it's not actually so simple. The clipboard is system-wide shared state -- what if another application changed the clipboard between the cut and the undo?

The other part of it ("ghosting" the cut) just seems like a matter of taste, and at some point you have to consider whether going against decades of established muscle memory in the users of your software is really worth it.

That said, it is a good idea to re-examine our assumptions every once in a while, so kudos for that.

This is getting a bit off-topic, but there are certainly people who will still say that they built it. That is, in a nutshell, how people try to reconcile desert theory with the existence of the super-wealthy.

Right. There is something interesting here that could be explored thoughtfully. The hard part is that we rely on compilers being correct, and they mostly are.

We have no viable mechanism yet to get the same level of confidence if some LLM-based system writes the binary.

Perhaps we can get to a system that produces not just the binary but also a machine-verifiable proof that the binary implements some higher-level language description of the program.

Though then the question will be whether we've gained anything, or whether we've just replaced the compiler with something massively more expensive that does the same thing.

There's some potential here for the LLM-based system to drive better performance optimizations than a regular compiler could.

Of course this isn't what Elon is actually saying, and we'd be better off if fewer people listened to him.

A dubious analogy.

Factories are places for the mass production of identical or nearly identical widgets.

There are some kinds of mass produced software, like the low value apps that lots of businesses want to have for some reason and that should have been websites instead.

But actual progress comes from software that isn't mass produced. So choose your ambitions wisely.

We can point out mistakes that feel rather grating without assuming intent behind them.

I agree that their use of "he" is likely because they're not a native speaker, especially because they're arguing against the capabilities of LLMs.

That doesn't make it inherently wrong to point out the mistake when it's so intertwined with the deeper discussion here, especially given the fact that some (hopefully few) people do build relationships with LLMs.

That's true, but it does mean that the LLM itself actually does have access to those thinking traces and could therefore, at least in principle, answer what it was thinking. They're probably not trained to do that, though.

It could potentially be falsified by an encounter with really weird aliens.

But I find it quite plausible because it feels like it's fundamentally "just" a restatement or a minor variation of the Church-Turing thesis.

To expand on this, I've found that often, the best way to make a change is in two steps: First, make the change easy (the re-architecture). Second, make the change that is now easy.

My best software development experiences have been in cultures where this way of working was common.

It is a shame that these cultures are not more widely spread; and I mostly blame GitHub: the lack of good support for stacked PRs or patch series or whatever you want to call it makes it harder than it should be to work in this way.

This doesn't actually seem to be true based on a quick googling, i.e. Germany has somewhat higher median income.

But in addition to the raw numbers, you have to keep in mind that they don't account for cost of living and that different countries account for various services differently, especially health care.

The trick is to do something else in those 20 minutes (or, ideally, even longer).

That's the main value I've been getting out of coding agents. I have them do (comparatively) simpler tasks or explorative tasks in the background while I'm in a meeting, doing code reviews, or otherwise working on something else.

That was probably not GP's point, but they can be. Sound-based weapons are a thing: https://en.wikipedia.org/wiki/Long-range_acoustic_device

Of course a lot of it is about the energy and overall exposure, and the harms of this, if any, are more likely elsewhere, but it's completely reasonable to question extraordinary promises made by people who up to this point have shown no expertise in the field.

I swear, it's like some people have already forgotten about Theranos.

Agree with most of this except come on. Facebook was born out of a misogynist precursor. The idea that its CEO had no malicious intentions is ridiculous.