That's okay. Just train another AI to look at the output of the first AI and rate it. Then, if the rating is too low, or if it breaks tests, roll it back. This is a necessary first step towards the AI learning to delete the tests before submitting like human coders do.

You forgot people who, rather than use AI, find ways to jam it into other people's workflows, so that those other teams get bogged down in slop, while the teams doing it get credit for expanding the use of AI. :-)

Why? I can count the number of times I've needed this as a software engineer on two hands when counting in unary. It's not that important a concept, and it's also not difficult.

Boolean logic is useful, but it is by far the simplest part of programming for people to understand. The part that students have a hard time doing, in my experience, is creating rigorously defined instructions with proper syntax, communicating code flow concepts like "do this while/until that", understanding data structures at a conceptual level, etc.

If I were trying to teach kids programming, I'd start with BASIC, then add data structures a couple of years later. Better yet, just teach them more abstractly how to give someone instructions, and teach skills that improve spatial reasoning. These will have a much bigger impact than understanding bit math or hex or boolean expression simplification.

Music classes, geocaching, and art are all more useful in terms of teaching the skills that make you a good programmer than any of the math concepts you're talking about.

We really don't need a large number of people working on silicon process or CPU design, and knowing how things work at that level is of almost no benefit to software engineers. It's way too low-level to matter. Even high-level CPU concepts like pipeline bubbles have at best marginal real-world utility, much less anything below that.

The number of times I have thought about flip-flops or ring counters while coding is exactly zero. The number of times I have thought about registers is exactly zero, because the compiler manages them for me, I have no control over them, and there are rarely enough registers for me to be able to meaningfully structure my code to minimize register fetches anyway, realistically, even if I could.

This is the point where your ideas start being useful, IMO. But on top of that, I think everyone needs a solid OS programming class where they have to implement virtual memory and drivers. Understanding how those pieces work is useful in understanding why software behaves the way that it does, particularly where performance is involved. Add a distributed computing class to force a rigorous understanding of synchronization and performance impact.

I would not assume Catholic education is right-biased.

What we're seeing are actually second-order effects of conservative policies towards education — specifically, chronically underpaying teachers. Teachers often don't know how to teach the material because they didn't really learn the material in the first place, so the blind are leading the blind.

This is what happens when you pay teachers poorly for decade after decade. Instead of getting the people most qualified on the material to teach it, you get whoever is willing to do the job at the sad salaries that they pay. So you have two choices: Teach the teachers or pay massively more so that you can steal people away from industry to become teachers. Those are your options.

Yeah, by comparing human energy consumption with AI energy consumption, this one seems to be arguing that their tech will replace human labor, and therefore, we won't need to feed the humans anymore.

You can't compare the energy spent training a model to the energy spent training a person, because unless you stop letting people procreate, you're not going to stop training the people. That's a committed cost. Future AI model training isn't.

Realistically, probably not. The kind of person who would rape and murder a 13-year-old girl almost certainly isn't thinking about whether he'll get caught.

And more broadly, there's no evidence that the death penalty makes murder less common, so if that's not enough to move the needle, then inadequate fear of getting caught likely doesn't play much of a role in crime rates.

This is very true.

I mean sure, there could be some kind of conspiracy, though Occam's razor suggests that it is more likely that the person who created it is doing something questionable.

On the other hand, given the current state of the U.S. government, the possibility that they are being targeted for having cached some kind of data that the Trump administration doesn't want seeing the light of day is not nearly as unlikely as one might hope.

Out of curiosity, I looked up Donald Trump on there just to see what it said. Ignoring a couple of really recent posts that seemed jarring by their inclusion, it mostly seemed like a competent summary of a lot of stuff, and I didn't see much there, if anything, that I disagreed with.

The problem was what *wasn't* there. The impeachments were about a paragraph long. Basically nothing about the January 6th committee was there, none of the myriad civil and criminal cases were there, etc. It was the most whitewashed article I've seen in a long time.

And then I looked up "Barack Obama". Far from being whitewashed, that article was mud-bathed. The controversy section was quite detailed, but tended to leave out any details that could be seen as positive. For example, it points out the 2013 Inspector General report about the IRS targeting conservative groups, but conveniently omits later revelations that at the time he made that report, he was aware that liberal groups were similarly targeted, thus creating a false narrative that the behavior was specifically anti-Republican, when in fact, it was anti-PAC. Still problematic, but very differently so.

So basically, the information that is presented seems to be mostly correct, but is massively right-biased by the information that it chooses to leave out. And lest you think that they just omitted things they didn't consider important, I would point out that both articles were packed with copious amounts of tedious and meaningless trivia that nobody should care about, and the articles could easily have been reduced by 90% without meaningful loss of useful information. For example, how many people care that one of Obama's elementary schools had a lot of Muslims in attendance? Exactly nobody... assuming it is even true and is not an AI hallucination or right-wing propaganda.

I'm not impressed. A trustworthy source should omit trivial and unimportant details, but should not omit details that would bias opinion. This reliably failed on both counts. Horribly. Irredeemably. To the folks working on that site, my advice would be this: Start over. This is trash.

But I guess that's the whole point — creating a right-biased Fakeipedia so that people don't have to be exposed to facts that exceed the limits of their political bubble world. So in that regard, I guess they succeeded. But I have exactly as much trust in that as I do in AI hallucinations.

And as I read this, I realize that the bill hasn't passed yet. So consider this a pledge that if my assembly member does vote for it, I will support his or her opponents, without regard to their positions on any issues, simply because almost anyone would be better than someone clueless enough about technology to vote for a law like this.

The California standard is completely and utterly infeasible. It fails Goodhart's law, which states that as soon as a measure becomes a target, it ceases to be a good measure.

As soon as you say "A part that looks like this arbitrary gun part cannot be manufactured", it's open season on designing a new version of the part that doesn't trip the rule. And as soon as the government adapts to that new design, five minutes later, there's another new design that still doesn't trip the rule.

It's a game of cat and mouse, but there are infinite numbers of ways to design the part, and only finite storage for models to recognize the part.

Basically, the game is already lost before the government even starts to play. The only way this would ever work would be if you were required to provide documentation telling why you're making each part, and what it is, and having someone validate that the part is plausibly what you say it is based on it looking like something you are allowed to print, and therefore presumably not part of a potential gun. And even that wouldn't be foolproof.

But as soon as the presumption is that it is okay to print something unless it matches any sort of pattern, model, etc., you're screwed. This can't be done, period, or at least not in any way that will meaningfully prevent... well... anything, really.

No, they won't. Nobody will care. Just like they don't care about all the murders that happen now. The only way somebody cares is if some blowhard politician screams, "Oh, my g*d! It's a ghost gun! Everyone has to panic and be outraged!" And then they will care. But as long as the people in power don't act like morons, someone will use a ghost gun, and there will be exactly as much of a response as for a murder involving any other firearm, because it really doesn't make a dime's worth of difference in the end.

Traceable guns don't catch murderers all that often. The overwhelming majority of gun crimes are committed with guns that were not purchased through a legitimate dealer — almost 86%. Even if the serial numbers have not been removed completely, you'd end up tracing the firearm back to the person from whom it was stolen, rather than the actual killer.

There is basically no benefit to serial numbers on firearms from a public safety perspective, and there never was. The only real benefit is getting stolen firearms back to their rightful owners. And while that's a laudable goal, the people pushing it as a way of preventing crime are being completely disingenuous.

The people pushing fear of ghost guns are doubly so. Every gun used in a crime is a ghost gun. The only difference is that someone of them were manufactured, and the rest were just de-serialized.

Anyway, my current plan is to figure out whether my assembly member voted for this trash, and if so, pledge the maximum dollar amount to the campaigns of every person running against that person in the next election. If everyone reading this post did the same, these shit-for-brains-stupid laws wouldn't keep getting passed.

Seems like after two days, the emergency services are going to come in and find a bad smell. Would be better to have an app for your watch that detects loss of consciousness/heart rhythm and calls emergency services when it would actually be early enough to maybe do some good.

The thing is, you don't even have to have collusion. When there are only a few companies in a space, there's no incentive to innovate, and there's no incentive to spend money that would drive operating costs down to compete with the other companies. Therefore, the cost stays high. Almost invariably, the appearance of a new major player drives costs suddenly down, because all of those cost-saving innovations suddenly are worth spending money on, because the new competitor jumps in with the latest, cheapest tech. But until that happens, even without collusion, there's no strong incentive for any of the companies to disrupt the status quo, so they don't.

False. For the most part, oligopolies don't compete meaningfully. There's usually zero price competition, often with most of the firms taking their cues on pricing from whichever company is the price leader for the market. They may differentiate their products slightly from one other, but not by enough to convince a significant number of customers to change providers.

But I don't want any DRM. Can't we get a panic screen without DRM?

Well you could always get DRM, DRM, Panic, Sausage, and DRM. That's only got a little bit of DRM in it.