Cars have software. But I don't think cars are software. Can I apply a software update to make my Honda Accord into Tesla or Dodge Ram?
> The only difference between a Tesla and an economy car from Stellantis is whether the software is well written or not.
Is that actually true? I mean, assume I have access to all software in the world and all IP lawyers got kidnapped by aliens - could I just write a software for Stellantis Economy to turn it into Tesla (or vice versa)? I don't think so.
> Cars have software. But I don't think cars are software. Can I apply a software update to make my Honda Accord into Tesla or Dodge Ram?
That's a disingenuously literal misinterpretation of what I said. I wasn't saying that a Tesla and some economy car are identical, only that they have in common the characteristic of being defined at their core by software. It should go without saying that software alone can't turn a Cherokee into a Model Y for the same reason that software alone can't turn a HomePod into an Apple Watch.
But there's an obvious difference between a good software experience and a poor one. Like in my wife's Cherokee, how the radio always turns on every time you start the car, no matter what you do. Like how the digital speedometer is completely concealed by any warning text that appears. Like how all window controls stop working as soon as any passenger opens their door after stopping the engine. This is all software, and I write this in response to rkagerer saying "no thank you" to cars getting meaningful software updates.
> I wasn't saying that a Tesla and some economy car are identical,
You literally said:
> The only difference between a Tesla and an economy car from Stellantis is whether the software is well written or not.
You didn't say "one of many differences". You said "the only difference". Maybe you wanted to say something else, and you still can, but you can't claim it's my fault you said that.
> It should go without saying that software alone can't turn a Cherokee into a Model Y for the same reason that software alone can't turn a HomePod into an Apple Watch.
Which invalidates your statements that the cars "are software". They are more than software. They are a complex combinations of software and hardware, each of them having its part - and, obviously, if one of the parts is bad, it makes the car worse.
I'm not sure what exactly pisses me off so much in this idea - after all, I am not upset by the existence of $Brand Basic, $Brand Premium, $Brand Luxury and $Brand Now-Everybody-Knows-You-Have-Money, each of which has different features and bells and whistles. But put it in one single box and charge me monthly rent to go from Basic to Premium - and it does feel wrong. Even if TCO of Premium comes out as lower over time. I don't know why exactly it feels that way but it looks like it feels that way to a lot of people. Maybe it's daily reminder that all the luxuries are right here, right under your fingers, if only you weren't so miserably poor? Or the constant necessity of begging somebody else for permission to use your own car (yes, car loans, but they feel different)? Not sure. But it feels like it's real, even if it's only in my head.
I think you've captured it perfectly with "Maybe it's daily reminder that all the luxuries are right here, right under your fingers, if only you weren't so miserably poor?"
Having age verification in every operating system? I think it is onerous. Imagine you need to update every embedded system because your wise lawmakers made it a crime to run any code that does not include age verification API.
"Behavior, graphics, etc." would likely constitute separate IP from the code. I am not sure there's a model that allows you to make AI reproduce Minecraft without telling it what "Minecraft" is - which would likely contaminate it with IP-protected information.
> Blanchard's account is that he never looked at the existing source code directly. He fed only the API and the test suite to Claude and asked it to reimplement the library from scratch.
I don't see how it matters what he looked at. If I took a copyrighted code and run it through a script that replaces all variable names, and then claimed copyright on the result because it's an entirely new work and I did not look on the original work, I'd be ridiculed and sued, and would lose that lawsuit. AI is a more complex machine, but still a machine. If you feed somebody'd work into a machine, what comes out is a derivative work.
Test suite is a part of copyrighted code, is it not? If he used just the API description, preferably from a copyright-clean source, then we could claim new work (regardless of how it was produced, by using Claude or trained pigeons or by consuming magic mushrooms). But once parts of the copyrighted code had been used, it becomes derivative work.
> AI is a more complex machine, but still a machine. If you feed somebody'd work into a machine, what comes out is a derivative work.
I'm not sure that's true, legally speaking. If you fed it into a PRNG, the output seems to me like it would not be an obviously derivative work (i doubt you could copyright it but that's a separate question). So we have 1 machine that can transform something into non-derivative work, and another that leaves the result derivative. The line isn't likely going to be drawn as "did a machine do it or not", but on a fuzzy human line of how close the output seems to be to the original (IANAL).
PRNG is not intended to use information sent into it in a substantive way (in other words, for PRNG does not matter if what you feed into is is Shakespeare's sonnets or white noise). Sure, if your machine is an electronic analogue of a shredder, then yes, the result is not a derivative work in any sensible meaning. But LLMs are not that kind of a machine.
I agree! But that's exactly my point: 1 type of machine is ok, another isn't, so it is not just a matter of all machines make derivative work. To draw the line more carefully is an open question. I would be surprised if a machine that "uses information sent into it in a substantive way" is a perfect deliminator. OTTOMH musicians might present some compelling objections.
> you should be aware of its whereabouts at any given moment
Says who? If the car is mine, I am free to do with it whatever I like (of course, excepting criminal acts). I do not owe anybody an account of what I - or the care - did at any particular moment. If the car was used in the commission of a crime, it's up to the prosecution to prove I had something to do with it. If they think I know who did it - prove it and prosecute me under the law. You can't just prosecute because you think I should know, that's not how proper law works - otherwise every cop in the country would be 100% sure who they caught is the criminal - because why not, if it's enough for conviction, why work harder!
> If someone shoots a person with your gun, you gonna say it wasn't you and expect them not to question any you further?
They can question all they like, but to secure a criminal conviction, they must prove, beyond reasonable doubt, that I was the person who did it. Otherwise you get no conviction. If they strongly suspect I did it, they would find a proof - but the fact that I owned a gun is not that proof (for one, guns can be easily stolen, and frequently are).
In Europe the law argues that cars are dangerous, and if you loan your car to a habitual bad driver, that's on you. You can either get the person who drove it to fess up, or the judge can fine you (because you lent out your car against better judgment) and impose a drivers log, so the circus doesn't happen again.
The arguing about having a constitutional right to drive bad boggles the mind, road deaths in the US are high, compared to civilized nations. Wikipedia states it's 14.2 deaths per 100000 inhbitants, that's between Sierra Leone (13.8) and Angola (15.0). For comparison, India has 12.6 traffic deaths per 100000 citizens and the worst country in Europe is Greece at 6.1.
The right metric is death per citizen, not per mile, because it's about the number of people who have lost a family member or friend.
When you get around exclusively on two wheels (motorcycle and bicycle) bad drivers are a direct safety threat. Even cagers ought to be careful about being permissive with red light running, side-on crashes are remarkably deadly for the one who got hit in the door because there is not much structural protection or space on the side of the vehicle.
But couldn't you then have the same argument for speeding tickets (or parking tickets)? Like, "I don't know who drove my car too fast or parked my car on the curb, so it's not my problem. The state should prove who did it.".
For the speeding ticket, if the police officer stops you, you'd get a ticket and would sign it, thus ensuring you were there. Or, if you refuse, the police officer would testify in court (if you would ask for a court hearing) that they saw you behind the wheel (and, likely now, you'd also be on bodycam). That would likely be enough evidence - which is why you probably don't want to go to court, because the judge would be annoyed at you for wasting everybody's time, and you'd probably get more severe punishment.
That said, I did at times get smaller fine and less severe consequences from a speeding ticket by just pretending I am going to go to court (I didn't really want to, I wanted smaller fine :) - because policemen do not like to waste time in court either - so they would agree, that if I do not try to deny I did it, and do not force thus them to go to court and testify, they would agree to less severe violation (while still costing me $$, just not as much as it could). That's totally a thing, at least in the US. The risk, of course, if you are an ass about it and piss off the police officer, they'd say to heck with it, I'll go to court, and you'd have to go to court too, and as per above, you'd get punished more severely. So, always be polite, and it will be to your benefit.
As for automated speeding tickets, I'm not a huge fan of it. Too many cases of this system being wrong or abusive.
I meant speeding cameras, where a system OCRs your license plate and you receive a fine after a month or even two months. Fine being monetary + points, depending on how fast you drove. Police officers measuring speed have become quite rare, at least here at my location (EU country). While speeding cameras are becoming quite common, especially in towns/cities and on the highways.
TBH, if those became legally invalid, I wouldn't lose a lot of sleep. There's a huge potential for abuse in these things, and their deterrent value is minimal. Especially in a heavily heterogenuos country like the US, where each municipality can have their own rules and their own courts. Sometimes these thing turn into literal highway robbery.
Yeah, keeping this would be a dangerous precedent. If the state can presume you're guilty in a traffic case, why not extend it to other cases? Stuff like that is routinely used in legal arguments, "we are doing X so why can't we do Y which is essentially the same?" So say they'd go for "we have your phone located within the vicinity of where murder is committed, now prove you're not a murderer!" or "your license place was tagged next to the store that was robbed, now prove you didn't rob the store!"
And yes, very likely some people would abuse it to get out of traffic tickets. I'd rather have that than constitutional due process protections eroded. We're not doing super-great on that anyway, we don't need to do worse, and if some scoundrel occasionally not paying traffic ticket is a price we have to pay to avoid that, I am fine with it.
Yup. Camel's noses should always be shot. Otherwise they creep in more and more.
Some examples that come to mind:
Look how the exception for searches at border crossings has expanded.
The use of actions against licenses for behavior that has nothing to do with the license.
The use of permits to get companies to do things only marginally related to the purpose of the permit.
The encouragement of universities to expel those accused of criminal acts--just because the punishment isn't jail should not mean the state can hand it off to a kangaroo court.
Pressuring financial companies to cut ties with disliked things. (For example, getting Steam to remove games with any whiff of incest. Either declare them illegal or don't take action against them!)
This article's numbers are (not) surprisingly useless. So, for 1.8% of people in one study who did MRI they found cancer. Is it a lot? A few? I mean, how many people do these MRIs, out of general population? Are those high-risk people or just random people (or, maybe, only rich people?) and what their base rate incidence of cancer? What is base rate incidence of cancer in general population and how often those cancers are discovered by non-MRI tests?
If I had to decide whether or not MRI is worth it for cancer detection - this is what the article is supposed to be about? - those are the questions I'd ask. None of them are addressed in the article. I think if we dropped everything in it but the last sentence, it would lose nothing of substance:
> No matter what you decide for yourself, consult a licensed medical professional for advice.
> if your downtime detection system kept producing false negatives, would your solution to be just turn it off
It's more like if your CI build fails in 95% of runs, and in only 2% of runs it indicates a real bug, do you have a bad CI which is next to useless for detecting bugs? I'd say yes, that's exactly what you have. No developer is going to pay any attention to this CI, and if you tell the developers they must ensure tests pass on such CI, they would rebel.
> The only difference between a Tesla and an economy car from Stellantis is whether the software is well written or not.
Is that actually true? I mean, assume I have access to all software in the world and all IP lawyers got kidnapped by aliens - could I just write a software for Stellantis Economy to turn it into Tesla (or vice versa)? I don't think so.
reply