22KW charger is now NLA

[daveo4EV]

Oh, what I like to call "the great compromise", a.k.a. design-by-committee. One of my top professional pet peeves. Don't get me started ranting on this topic.

some day when I'm really really bored I _WILL_ get you started - I'm strangely interested to see how much of your rant we share

Sponsored

 

[dropdead]

Your comment on entry level however has a flip side of a coin. How are we going to get experienced engineers if entry level positions are replaced by AI. Is AI going to "grow" into experienced engineers?

Good catch, sir! The issue is, you don't. The kids coming to the job market now are cooked, as they say. If you are lucky and actually survive the AI-based HR-interview-loop for the few jobs that software company still hire for, it is usually expected that you do vibe-coding and are a 100x engineer that basically just tells agents to do stuff instead of actually solving those problems and building good architecture.

chatgpt and the other LLMs are as good as it's gonna get since late last year. The point of diminishing returns already has been reached. We are in the stage of "agents" now, where boards and CEOs push companies towards building "virtual engineers", "virtual supervisor", "virtual boss", force team leads to only think about KPIs like burndownrate (aka how much tickets are done by a team in a given interval) - forcing assumptions like "i heard salesforces increased it's rate by 60% since last year, why is your team only have an increased rate of 20%, you definitely need to enforce that they use AI more" - that is currently one of my main jobs as interim cto, to manage these kinds of expectations...i am getting paid good money for that but it has been hard on my motivation for good engineering.

 

[whitex]

Good catch, sir! The issue is, you don't. The kids coming to the job market now are cooked, as they say. If you are lucky and actually survive the AI-based HR-interview-loop for the few jobs that software company still hire for, it is usually expected that you do vibe-coding and are a 100x engineer that basically just tells agents to do stuff instead of actually solving those problems and building good architecture.

So again I ask, what is the expectation of those companies what to do in 20 years when the experienced engineers retire and all they got left is today's gen Z with 20 years of nothing but vibe coding experience, now flying without the older engineers who still remembered how to code without AI?

 

[whitex]

chatgpt and the other LLMs are as good as it's gonna get since late last year. The point of diminishing returns already has been reached. We are in the stage of "agents" now, where boards and CEOs push companies towards building "virtual engineers", "virtual supervisor", "virtual boss", force team leads to only think about KPIs like burndownrate (aka how much tickets are done by a team in a given interval) - forcing assumptions like "i heard salesforces increased it's rate by 60% since last year, why is your team only have an increased rate of 20%, you definitely need to enforce that they use AI more" - that is currently one of my main jobs as interim cto, to manage these kinds of expectations...i am getting paid good money for that but it has been hard on my motivation for good engineering.

The only thing this will help grow is hardware. Hardware running all the AI model training, hardware running all the AI agents, and then hardware which will run all the highly bloated AI generated code:
Vibe coder:
write code to blink an LED
AI agent :
here is the code, here are all of its dependencies, and it requires a system running Linux with a minimum of 2 ARM cores (4+ recommended), min 2GB or RAM, 16GB recommended). Would you like to to add a web services client code so that the LED can be controlled from your phone anywhere on the internet? If yes, I will write you code for your cloud instantiated virtual servers and the phone app for iOS and Android.
Vibe coder:
Bet! That will make an slaps demo to my leadership!
AI agent:
Please cleanup or upgrade your hard drive space, this code requires a minimum 2TB of disk space to build.

2 days later
Vibe coder:
The LED is not blinking after an hour.
AI agent:
You are right, let me see if I can fix it for you. Here you go.
Vide coder:
It's still dying after an hour or so.
AI agent:
You are right, let me see if I can fix it for you. Here you go.

A week later:
Vibe coder to his AI virtual manager:
I've been debugging for days now, I think I need a next generation AI model subscription.

 

[dropdead]

The only thing this will help grow is hardware. Hardware running all the AI model training, hardware running all the AI agents, and then hardware which will run all the highly bloated AI generated code:
Vibe coder:
write code to blink an LED
AI agent :
here is the code, here are all of its dependencies, and it requires a system running Linux with a minimum of 2 ARM cores (4+ recommended), min 2GB or RAM, 16GB recommended). Would you like to to add a web services client code so that the LED can be controlled from your phone anywhere on the internet? If yes, I will write you code for your cloud instantiated virtual servers and the phone app for iOS and Android.
Vibe coder:
Bet! That will make an slaps demo to my leadership!
AI agent:
Please cleanup or upgrade your hard drive space, this code requires a minimum 2TB of disk space to build.

2 days later
Vibe coder:
The LED is not blinking after an hour.
AI agent:
You are right, let me see if I can fix it for you. Here you go.
Vide coder:
It's still dying after an hour or so.
AI agent:
You are right, let me see if I can fix it for you. Here you go.

A week later:
Vibe coder to his AI virtual manager:
I've been debugging for weeks, I think I need a next generation AI model subscription.

yea that sounds about right, happening already as we speak

 

[Brombaer1971]

Vibe coding as everything else which is KI currently experiences the hype cycle with all its easy victories and hard problems. I do not see it as a threat, it is just another tool in the hand of experienced developers which will deliver but not in the way that it does big things. What @whitex is describing sounds to me a bit like the windows time with activex components written by inexperienced people being thrown together by other people without deep thought.

Currently there is backslash again vibe coding because it does not deliver as expected (overhyped). So I am more calm, just waiting for the future.

How Porsche does software development, testing and releasing to the Customer I don‘t fully understand. For the Customer it is clearly visible that there is a lot that could be improved. And as a developer I find it really hard when someone tells me that the software needs to be updated at a dealer and needing hours which results Porsche in being reluctant to role out every improvement avoiding the huge update costs. In the end its the customer that suffers by this decision.

 

[daveo4EV]

Vibe coding as everything else which is KI currently experiences the hype cycle with all its easy victories and hard problems. I do not see it as a thread, it is just another tool in the hand of experienced developers which will deliver but not in the way that it does big things. What @whitex is describing sounds to me a bit like the windows time with activex components written by inexperienced people being thrown together by other people without deep thought.

Currently there is backslash again vibe coding because it does not deliver as expected (overhyped). So I am more calm, just waiting for the future.

How Porsche does software development, testing and releasing to the Customer I don‘t fully understand. For the Customer it is clearly visible that there is a lot that could be improved. And as a developer I find it really hard when someone tells me that the software needs to be updated at a dealer and needing hours which results Porsche in being reluctant to role out every improvement avoiding the huge update costs. In the end its the customer that suffers by this decision.

one of the "core" problems is that the vehicle's lack a 12V system that has enough capacity to run all the systems while the big battery is offline during an update - one of the main parts of doing the software updates at the dealership is attaching the vehicle and it's 12V system to a known stable/reliable 12V external power supply - which will power the vehicle during the extensive (multi-hour) update - there are also complex, multi-step software patches that must be applied in the correct order (or you'll brick the vehicle or expensive subcomponents that can only be replaced once bricked - not recovered)…

I watched a software update once for 6+ hours - and yeah - the 12V battery would've gone dead during the update if there was no external 12V power supply…

this is not the only problem with software updates - but it's a major contributing factor…

 

[whitex]

one of the "core" problems is that the vehicle's lack a 12V system that has enough capacity to run all the systems while the big battery is offline during an update - one of the main parts of doing the software updates at the dealership is attaching the vehicle and it's 12V system to a known stable/reliable 12V external power supply - which will power the vehicle during the extensive (multi-hour) update - there are also complex, multi-step software patches that must be applied in the correct order (or you'll brick the vehicle or expensive subcomponents that can only be replaced once bricked - not recovered)…

I watched a software update once for 6+ hours - and yeah - the 12V battery would've gone dead during the update if there was no external 12V power supply…

this is not the only problem with software updates - but it's a major contributing factor…

That problem is only there because of two reasons:

1. Porsche hasn’t invested enough effort into mandating faster update times for the various ECU’s. No update should take 6hrs.
2. Porsche is trying to keep their vehicle software architectures the same between ICE and EV platforms. For EV’s, they can simply turn on the DC-2-DC converter and the HV battery will keep the electronics going for way longer than 6hrs (provided HV SOC is at some minimum)

They didn’t even have to invent it themselves, Tesla has had OTA updates since MissionE days, they could have just copied the behavior. Trying to reuse ICE parts and designs for EVs is holding back traditional auto manufacturers. It would be like a horse and buggy manufacturer trying to reuse parts for an ICE car.

 

[Murph7355]

So again I ask, what is the expectation of those companies what to do in 20 years when the experienced engineers retire and all they got left is today's gen Z with 20 years of nothing but vibe coding experience, now flying without the older engineers who still remembered how to code without AI?

I don't disagree with the general theme, and think great care is required in how we deploy AI and what we use it for... I studied it in 1993 and half of the course was around the philosophy of it... "just because we can, should we?".

The but...

How many engineers are out there now that understand machine code/assembler/etc? (Am assuming it's not many). And how have those numbers impacted the industry?

Things will ebb and flow. We will hit edge cases that AI cannot be used for and there will be those who gravitate to being able to resolve them. Initially they'll do this because they are inquisitive, but money will then draw in more. And if AI then develops to solve the edge case (because humans become expensive enough to justify the development) then things will flow in a different direction.

We evolve. When the internal combustion engine was developed there were those noting "but surely we can't work without horses. What happens when nobody re.mbers how to use horses and ICE fails"... Or when automation came to mills, "what if we forget how to do it by hand". Over a century or two later, neither has been a problem. Nor will AI be. (Until it becomes sentient and realises we're surplus to requirements ).

 

[Murph7355]

That problem is only there because of two reasons:

1. Porsche hasn’t invested enough effort into mandating faster update times for the various ECU’s. No update should take 6hrs.
2. Porsche is trying to keep their vehicle software architectures the same between ICE and EV platforms. For EV’s, they can simply turn on the DC-2-DC converter and the HV battery will keep the electronics going for way longer than 6hrs (provided HV SOC is at some minimum)

They didn’t even have to invent it themselves, Tesla has had OTA updates since MissionE days, they could have just copied the behavior. Trying to reuse ICE parts and designs for EVs is holding back traditional auto manufacturers. It would be like a horse and buggy manufacturer trying to reuse parts for an ICE car.

I'm not convinced that's the issue. It feels more like a fundamentally poor architecture that's far from ideal no matter what the powertrain. And rather than fix it properly they just layered on poor fix after poor fix (“another ECU will fix that". Why so many feckin ECUs!?).

Maybe the Macan EV system is aiming to address that, but I doubt it. Porsche has excelled at mechanical engineering (mostly) but is piss poor at computer architecture and too arrogant to realise it (why do they have their own division within VAG? Your VW system is shit...hold my beer).

The "keeping the car plugged in" angle is a non-issue. Easily solved. But solving 6hr updates means a full return to the entire computer architecture drawing board.

 

[BjörnfromHamburg]

We got waaaay off-topic in this thread.
Both topics, the original one and the one being discussed the last few pages, are interesting.

Any possibility of splitting this thread into two?

 

[Brombaer1971]

I fully agree with @BjörnfromHamburg.

I usuallly come back to say, we as a user do not really understand it (and shouldn‘t), it is the full responsibility of the manufacturer to get it done. They are projecting their „luxury/sporty“ brand image on us, but leave us to deal with the ugly things. There is no reason to not fix things that are bad for the customer. And just because things have been not good in the past doesn‘t mean they are to stay til forever. Plain and simple.

 

[whitex]

How many engineers are out there now that understand machine code/assembler/etc? (Am assuming it's not many). And how have those numbers impacted the industry?

Well, it’s not the understanding of machine code. It’s understanding computer engineering design. How has this impacted the industry, I think I will suggest you reading the many posts here on the J1.1 battery charging user interface, or the Porsche Connect service, as examples relevant to this forum.

My point on AI replacing junior engineers still stands. Most for profit companies will take the short term profit by using AI instead of junior engineers, assuming someone else will train them so they can be hired later. We could end up in a situation similar to air traffic controller shortage in the US - nobody had the foresight that the old guys will retire and that it takes years to train replacements.

 

[whitex]

And rather than fix it properly they just layered on poor fix after poor fix (“another ECU will fix that". Why so many feckin ECUs!?).

So many feckin ECUs is a design choice. You can achieve guaranteed non-interference much easier for example. Even Tesla has split off some ECU’s, despite their original plan to put everything in the infotainment ECU - infotainment today does a lot less “car” functionality than the original Model S infotainment did. They were running Linux on it, which did not allow them to do real time or guarantee non-interference. Heck, even with an RTOS I can still give you scenarios where interference between different software can happen. Extreme example, imagine a multi-core SoC, fully partitioned so they your critical functions run on their own cores, fully isolated as far as privilege levels, mapped memory, scheduling, etc. Now, what if someone runs some malicious loop like BurnCortex virus on a completely different core. At first glance you’d think no problem, different core, so at most that different core goes down. However, that one non-critical core now starts heating the entire chip, which can cause denial of service via thermal shutdown or throttling. Validation also gets much complex with different functions sharing compute resources. With separate ECU’s, there is physical isolation, as well as much easier validation. There are benefits to an integrated design too, but in the end it’s a game of trade-offs.

 

[whitex]

Maybe the Macan EV system is aiming to address that, but I doubt it.

From what I heard from Porsche tech who work my Taycan, the Macan EV is much more integrated. One of the things they were commenting on was the fact that because a lot functions are now unified in one ECU, if any function fails, they have to replace this uber ECU, which apparently is not cheap, effectively increasing repair costs for the Macan EV.

Sponsored