Hacker News Reader

Comment by dreamcompiler | Hacker News Reader

I expect they don't want to pay what great software engineers cost, and even if they did, most such engineers probably wouldn't want to put up with GM's bureaucracy.

It's probably kinda the same story with great mechanical and industrial engineers, which is why approximately nobody finds GM cars exciting.

View on HN

The tragic aspect is that they spend inordinate amounts of money on this software with nothing to show for it. The money is spent so poorly and ineffectively, while trying to cut every inconsequential cost corner along the way with the output being micromanaged by people that don’t understand software, that if you directed a fraction of that money at a small team of highly paid and experienced software engineers you’d likely get a brilliant product.

Automotive OEMs don’t work that way, they spray money at hundreds of the cheapest engineers they can get away with and hope something good comes out the other side. I’ve seen it over and over. Their problems could be solved by paying a small team of excellent engineers proper Silicon Valley wages. It would be cheaper and produce a better product, but they don’t think that way. The concept of a highly paid software engineer is objectionable to the executives in that industry.

Many years ago, I discussed possible employment with a major head unit manufacturer. They were very excited about modernizing but basically stunned into silence when I mentioned typical senior FAANG pay. I'd be surprised if things have changed much.

Sitting here in Michigan I still see auto company software positions sometimes that are mid/late career positions and pay under 100k. The legacy auto industry really doesn't get it. And now GM is forcing RTO, so I expect they'll lose more talent and have a harder time attracting engineers.

It doesn't take much more than not going for the cheapest offshore team possible to make better car media center code than we get. I'm sure most of us could easily do it ourselves in a month or less. It seems like such a simple problem. Play songs and interface with some controls/radio receiver/etc while being at least a tiny bit conscious of efficiency.

This is an incredibly myopic take bordering on arrogance. The infotainment system of a vehicle needs (as in, is mandated) to be as safety compliant as the rest of the vehicle's software so your steering doesn't lock out when the display crashes. Writing software for safety critical systems is not a task most developers are familiar with.

Uh, what? If my infotainment system is capable of influencing my steering or any safety-critical function then malpractice has been committed. Infotainment needs only read (one-way) capability on the vehicle.

I do wonder about the CAN bus. there's no access control on the CAN network, there's one shared bus for the whole car, and any controller can starve the bus by spamming with a high address, unless that's changed.

you'd hope that there'd be some kind of filtering to prevent the infotainment system from sending (arbitrary) CAN messages, but I recall some crazy demos of researchers pwning a car's accelerator after rooting the center console.

The Tesla Cybertruck doesn’t use CAN for steer-by-wire. It uses Gigabit Ethernet. According to my mechanic my ABS system is using something similar. CAN isn’t the only bus in the car. According to Wikipedia the LIN bus is intended to supplement CAN with non-critical components.

Most modern cars have more than one CAN bus.

(And it’s lower ID messages which have priority over higher, which is more trivia than argument against your premise.)

Is that really a security issue, though? If someone has access to your center console, they have access to your car. If someone has access to your car, they can cut your brake lines or do a million other things that are impossible to defend against.

The concern is that the attack could be carried out remotely.

https://www.wired.com/2015/07/hackers-remotely-kill-jeep-hig... is one such example (might be the one GP is referring to).

I agree that is a serious concern, then. I thought it sounded like voting machine hacking where the dramatic headlines are hiding the fact that the hackers had physical access to the machine.

Voting machines are scary for different reasons.

1) The general public has private physical access by design.

2) The chain of custody is unclear and hard to prove. This undermines trust in the system.

That depends on the jurisdiction. When I voted in Maryland, there was a machine that I directly cast my vote into.

When I vote where I live now in Massachusetts, I fill out a paper form in private and then I manually feed the forms into a voting machine where I have only supervised physical access.

I think it's entirely possible to have machine-counted ballots without giving unfettered and unsupervised access to the counting machines and I prefer the MA system (for the reasons you describe), but I also recognize the Constitution is clear in its reserving the power to each state for how to conduct the election (at least for President) in Article 2, Section 1, Clause 2.

Counting machines are different than voting machines.

Here in Washington we vote on paper but the results are counted by machine. The counting process is overseen by representatives of both parties. In a recount the paper votes could be manually counted to reconcile with the machine. I’m not worried about any of that.

What is scary about voting machines is that the output isn’t guaranteed to match the input. This is for any number of reasons. The best defense against this is to print the choices in a human-readable format… on a piece of paper. Which means you don’t really need the machine at all. If it’s on the network you can’t trust anything.

This is not correct. The main infotainment system is not considered safety critical. There are some functions such as rear view camera that are but often this is handled by a real time os that overlays the video on top of the infotainment output. The infotainment system crashing will not take down your steering or braking.

Can confirm that. My xdrive unit in my BMW died a few months after buying, and everything worked, I could even hear the parking sensors even if I couldn’t see the camera feed anymore because that was piped through the X-drive unit.

I have heard from industry insiders that this is true for some major manufacturers, and not for others. If the infotainment is integrated with the driving functions then every release of the infotainment system needs recertifying. Some manufacturers avoid this by having completely separate systems.

Dear no, rather exactly the opposite. In fact, I wouldn't be surprised at this point if all new cars have the capability to reboot the display system in the middle of driving.

You definitely don't want these systems intertwined. It has the side benefit of allowing you to do pretty aggressive things like using more modern hardware, software, etc. E.G. last time I checked, Tesla uses Godot for a decent amount of their UI.

Absolutely wrong. Infotainment systems in cars these days typically run on Linux, not a hard realtime OS, and are definitely not any kind of safety-critical system whatsoever. There's no need for it either; they're isolated from critical systems on the car, in case there's a problem (and there frequently is).

I don't think that's right. VAG's software is terrible and crashes all the time. I've been locked out of adjusting my AC on a long drive because the infotainment system crashed.

The problem is that you need to architect, build, and maintain that system for years, or decades, to come. That requires an in-house software team that knows what it's doing. Building that first version that "works" is easy. Building that first version that actually works and supports everything you need it to, is quite a bit more difficult.

Building and then maintaining that software, year after year, without slowing to a crawl because of all the tech debt that's piled on (especially so since you have subpar engineers or you're contracting out to cheap workers abroad), is even harder.

And if you want to provide a user experience comparable to what the top software companies in the world are capable of, that's just not gonna happen at these car companies.

Tesla was born from the software industry, that's why they're the only ones that can compete.

Uh… cars don’t really change from year to year. Pretending that an acceptable infotainment system is hard to do, just isn’t so.

The real problem is dumbass “features” (my MDX likes to show an animation of the keyboard typing out all the words a half second a letter, whenever you click autocomplete… for some reason) The real issue is that the infotainment systems are underpowered, because they’re considered an afterthought.

> infotainment systems are underpowered, because they’re considered an afterthought.

Also, cars sell in the millions. Car manufacturers will redesign a circuit board if it will save them cents on each board.

So it is always a struggle to put a more powerful CPU in there.

Infotainment responsiveness needs to be a NHTSA regulation. The mental load of dealing with laggy and unresponsive radios is surely causing accidents. Even if at worst it adds another $100 to the cost of the car to get a good enough cpu, so what?

Apple made a computer with a fast, responsive UI in 1984 using a chip that ran at 7.8 mhz and had 128 kb of ram.

You don't need a fast chip to eliminate lag. It's just incompetence.

Sure. I've used home computers since 1979. The graphics were instantaneous back then even on a few MHz and 16 kBytes of RAM.

We don't want slow and laggy UIs because the software is bloated.

But I responded to this part:

> The real issue is that the infotainment systems are underpowered

And my point was that it is far harder than it seems to put more powerful CPUs in cars.

They weren't doing that much either. I doubt an Apple ][ could even stream modern audio codecs via bluetooth, let alone simultaneously run a local search on the GPS database.

A friend of mine has a 2022 Cadillac something. I was actually very impressed with it. Comfortable, quiet, no-nonsense self-driving capability. Seemed like a fine car. If GM can build similar no-nonsense infotainment software they will be fine.

A no nonsense infotainment system doesn’t sound like the kind of thing people will spend $25billion a year on.

Good for them? Why should a good infotainment system cost $25B/year in perpetuity? Doesn't anyone solve problems anymore?

From the article: they want a slice of the $25 billion subscriptions.

Sure but who wants that? GM could be right that people want a car that hasn’t been enshittified.

The problem with chasing $25bn in subscription fees is you don’t know the opportunity cost of alienating your customer base.

> GM alone hoping to make as much as $25 billion per year just off subscriptions by 2030.

They think they’ll make $25B themselves. They also want your data.

Oh, yeah, that’s worse.

Why did mfr's GPS cost thousands as an option or subscription for 15 years while Garmin Nuvi portable devices for three hundred dollars ran circles around built-in units?

>I expect they don't want to pay what great software engineers cost, and even if they did, most such engineers probably wouldn't want to put up with GM's bureaucracy.

IMO, the bureaucracy isn't that big a problem; all big companies have this problem. The big problem for GM is their locations: who the heck wants to move to Detroit? Or any of the other places their suppliers tend to locate?

Super low cost of living on a tech salary? Sounds good to me. But if that doesn't work GM has the Cruise offices in San Francisco that they'll have to find some use for. GM proper also has offices in North Hollywood and Pasadena. Cadillac famously used to have offices in New York for the prestige, but Cruise still does AFAIK.

I'd bet that the problem is absolutely going to be more of a bureaucracy and salary issue than a location one. GM is a hilariously broken company and I can't imagine an enjoyable one to work for.

I was talking to folks who did this in the southeast. Compensation was the problem, not location.

For those folks, sure. For many others, the location could very well be. You won't know because many people probably only look at job ads filtered by location, so they just ignore any that are in "undesirable" places, whatever those may be.

GM has many development centers, including Roswell, Georgia, but it should be pointed out that like most American metros, Metro Detroit has a wide variety of cities and towns for different lifestyles. You don't have to live in Robocop to work for GM, which has facilities spread about both the Detroit metro area and the country.

And since software is built in big factories just like cars are, there is absolutely no way for those engineers to build those systems without moving to Detroit.

Reply on news.ycombinator.com