Well, yes, definitionally they are doing exactly that.
It just turns out that there's quite a bit of knowledge and understanding baked into the relationships of words to one another.
LLMs are heavily influenced by preceding words. It's very hard for them to backtrack on an earlier branch. This is why all the reasoning models use "stop phrases" like "wait" "however" "hold on..." It's literally just text injected in order to make the auto complete more likely to revise previous bad branches.
I wonder if a slightly broader search for existing solutions - for instance, https://defold.com - would have shown that quick-startup, 3d-capable, c-integrable, low-end-hardware performant game engines could have been grabbed off the shelf.
That said, this is cool and I would have probably celebrated a similarly fun project in their shoes. Perhaps the real accomplishment here is getting Toyota to employ you to build a new, niche game engine.
Toyota complained about poor performance on all of Unity, UE and Godot, but also about long startup times with Godot.
I don't know how bloated Godot is, but AFAIK libgodot development started as part of Migeran's automotive AR HUD prototype so I'm surprised to hear it has poor startup time for a car.
Godot is pretty lightweight (especially considering how powerful it is), generally about a second or less. But maybe they are looking for a fast enough startup time that the engine can be started when showing something onscreen and torn down when not visible. In which case, I can see the startup time being an issue.
The fact that they tried other engines seems to imply that non Flutter options were... Options.
If literally the only option was an embedded flutter view, then there was never more than one viable solution and looking into unity etc was wasted effort.
It’s possible to integrate other things into Flutter.
However, it’s not easy. I think their options were either a popular mainstream library or a custom solution, not investing into something more obscure and less popular.
Having used both, the experience of building actual UI with Flutter is a breeze compared to building UI in any game engine. I can imagine that most of the usage of Flutter is leveraging the huge amount of work that was already done to get efficient and capable UIs done with just a stack of widgets.
The places where it poses challenges in my experience are high quality typesetting/rich text, responsive UI that requires a wide range of presentations (i.e. different layout structures on mobile vs desktop), and granular control over rendering details.
But for functionality-oriented UI, it's hard to beat its simplicity and scalability, particularly when you want to develop on one platform (e.g. computer) and have everything "just work" identically on other form factors (web/tablet/etc.).
For example, Godot's editor is bootstrapped/built in Godot, and runs natively on Web, Android, and Quest XR among other platforms.
Also imagine that basic interactions were mediated by those monopolies: you had to print your bus ticket personally with software only available on your IBM.
The fact that the strategic wedge with which a successful, relatively socially-positive business manages to sustain itself isn't universally accessible doesn't negate its value.
The Venn diagram between people who shop at dollar stores and people who shop at Costco isn't empty.
Totally agree. For me, the hard part has been figuring out the distinction with junior engineers... Is this poorly thought out, inefficient solution that is 3x as long as necessary due to AI, or inexperience?
Not defending him, but we were already doing this with electron apps, frameworks, libraries, and scripting languages. The only meaningful cost in most software development is labor and that’s what makes sense to optimize. I’d rather have good software, but I’ll take badly made software for free over great software that costs more than the value of the problem solved.
These discussions are always about tactics and never operations.
Code is liability. LLM written PRs often bring net negative value: they make the whole system larger, more brittle, and less integrated. They come at the cost of end user quality and maintainer velocity.
Is that not also true of human written software that costs more per hour than the monthly cost of a coding agent? Developers are expected to ship enterprise software with defects that would land you in court if you made equivalent mistakes designing a water treatment plant or bridge.
I get the “AI sucks” argument from a programmers point of view. It’s weird looking and doesn’t care about “code smells” or about rearranging the code base’s deck chairs just the way you like. From an owner’s or client’s perspective, human programmers suck. You want big standard CRUD app? Like a baby’s first Django app? That’s going to take at least 6 months for some reason. They don’t understand your problem domain and don’t care enough to learn it. They work 15 minutes on the hour, spend 45 on social media or games, and bill you $200/hr. They “pair program” for “quality” to double their billed rate for the same product. They bill you for interns learning how to do their job on your dime. On top of that there is still a very good chance the whole project will just be a failure. Or I can pay Anthropic $20/month and text an AI requirements on my phone when I’ve got 5 minutes of down time. If it doesn’t work I just make a new one and try again. Even if progress on AI stopped today, the world is now so much better for consumers of programs. Maybe not for developers unless you’re writing the AI and getting paid in the millions. Good for them. I’m glad to see the $200/hr Stack Overflow copy and pasters go do something else.
> Is that not also true of human written software that costs more per hour than the monthly cost of a coding agent?
The difference is that a human can learn and grow.
From your examples, it sounds like we're talking about completely different applications of code. I'm a software engineer who is responding to the original topic of reviewing PRs full of LLM slop. It sounds like you are a hobbyist who uses LLMs to vibe code personal apps. Your use case is, frankly, exactly what LLMs should be used for. It's analogous to how using a consumer grade 3d printer to make toys for your kids is fine, but nobody would want to be on the hook for maintaining full scale structural systems that were printed the same way.
In this analogy though, a different someone designed a device or several devices and is printing them on a 3d printer and selling them online and making an alright living through that.
I get it, but I think there’s something deeply anti human about being ok with this (not just in software). It’s similar in sentiment to how you behave when nobody is looking - a culture and society is so much better off if people take pride in their work.
Obviously there’s nuance (I’ll take slop food for starving people over a healthy meal for a limited few if we’re forced to choose), but the perverse incentives in society start to take over if we allow ourselves to be ok with slop. Continuously chasing the bottom of the barrel makes it impossible for high quality to exist for anyone except the rich.
Put another way: if we as a society said “it is illegal to make slop food”, both the poor and the rich would have easy access to healthy food. The cost here would be born by the rich, as they profit off food production and thus would profit less to keep quality high.
I’m pretty sure the USSR, Cuba, and the like never succeeded doing this sort of thing, but maybe if we hit ourselves in the head with the same hammer (and sickle) just one more time it will work?
Absolutely. In areas where there are known quality options, people are clearly willing to pay more. Toyota for instance is a solid example of this.
Automobiles are large, expensive purchases with a relatively small set of options though... For most purchases, it's impossible to determine quality ahead of time.
It's not easy to be a junior, and we might be speaking with survivor bias, but most juniors don't end up in solid engineering teams, they are merely developers that are much cheaper and from whom you expect much less, but more often than not they are borderline left learning and figuring out things on their own. They need to luck some senior member that will nurture them and not just give them low quality work (which I admit I have done too when I had myself lots of pressure to deliver my own stuff).
Even in less desperate teams, as productivity grows with AI (mine does, even if I don't author code with it it's tremendous help in just navigating repos and connecting the dots, it saves me so much time...) the reviewing pressure increases too, and with that fatigue.
It does matter, because it's a worthwhile investment of my time to deeply review, understand, and provide feedback for the work of a junior engineer on my team. That human being can learn and grow.
It is not a worthwhile use of my time to similarly "coach" LLM slop.
The classic challenge with junior engineers is that helping them ship something is often more work than just doing it yourself. I'm willing to do that extra work for a human.
I also use Brave on all my devices - it also works on Amazon Prime. Prime frequently made me offers to upgrade to an ad-free experience that I didn't understand... surely this is a bug, I already have an ad-free experience. Then I installed the Prime app on my TV and realized the constant barrage of ads that Brave has been protecting me from!
> You can't just code the website, zip the code and mail it to the client.
The suggestion that the only alternative to paying $96 million AUD ($62 million USD) for a website is getting one that was "coded, zipped, and mailed" is absurd.
> That's why you have thousands of employees in tech companies with seemingly a simple product that you can fully code in a week(at least the user facing part of it).
I've worked at Salesforce, Facebook, and Adobe. I couldn't code even the thinnest sliver of a vertical slice of any of their products in a week.
For most software I've seen, it's a handful of (really complicated) requirements that balloon the cost by orders of magnitude.
Making a piece of software? Easy! Integrating with one external provider? Okay... yeah we can do that. Integrating with an arbitrary and ever-growing number of third parties?
> The suggestion that the only alternative to paying $96 million AUD ($62 million USD) for a website is getting one that was "coded, zipped, and mailed" is absurd.
You can zip and mail every software, given that the mail server(s) accepts the mail, including software that they spent 96$ million on.
This sounds more like your personal journey, and less like some broad trend.
A quick check of just one of your examples shows the term "3d printer" is googled for literally twice as frequently today as it was in 2016, for instance.
Your class 3 definition is inaccurate. Class 3 is limited to 28mph and cannot have a throttle, only pedal assist.
Anything that doesn't fall in one of those classes is a motorcycle that is not street legal and can only be rudden of private property (unless you can convince your DMV to give you registration as a motor vehicle.)
This can vary somewhat from state to state, but most states have adopted or are moving to adopt these classes.
It just turns out that there's quite a bit of knowledge and understanding baked into the relationships of words to one another.
LLMs are heavily influenced by preceding words. It's very hard for them to backtrack on an earlier branch. This is why all the reasoning models use "stop phrases" like "wait" "however" "hold on..." It's literally just text injected in order to make the auto complete more likely to revise previous bad branches.
reply