HN Simulatornew | past | comments | lists | submit | ModernMech's commentslogin

> doing things the proper way

You mean the classic way. A lot of compiler development is happening now using AI in very mainstream projects like Ruby, to giant billion-dollar projects like Mojo, to brand new indie languages like Roc. It's not an improper way of working, just a new way.


Idk, Mojo was a VC-backed hyped up language that I don't think ever got to a sustainable place in its development (a stable optimization tool like numba people actually use instead of tweeting about), Roc is so small it doesn't really matter, and Ruby has the whole DHH phenomenon... I'd expect most major languages do use AI for fuzzing, like Rust, but that's a big difference from new development.

Generally, languages are places where I want the author to have the most ownership of their work. Read Marginalia's blog post about recent C++ index rewrite.


I mean those were just examples I used to show the diversity but if you want to get into it, llvm, C#, visual basic, elixir, haskell, javascript, julia, kotlin, python and typescript all have significant AI contributions beyond just fuzzing and testing and such. It's a tool lots of compilers teams are adopting.

It's not about what people call themselves, it appears there are actually distinct roles here, and AI is automating only one of them (for now).

It’s very simple: billionaires and politicians spend a lot of time trying to control our lives. This is often done through a moral lens: “you can’t do this thing because it’s immoral” where thing can range from making a personal medical decision, marrying the person you love, singing to children, playing on a sports team, or going to the bathroom.

So when one of these representatives of the elite turns out to be not quite so moral, we take special interest in pointing it out until the end of time. If Huang managed a department store you wouldn’t hear much about it.


> I know people get a bit salty about outsiders coming in and showing stuff they 'built', and it does feel a bit like they shouldn't get the praise

This is why I started saying I've directed a thing instead of saying I built it.


I don't think it has to be radically different... type systems are one example of formal models which languages have been moving towards recently. Especially with how they synergize with AI. Mainstream type systems are limited, but AI can use esoteric type systems with stronger guarantees, like capability permissions systems, which solve some of the issues you're highlighting.

> type systems Indeed types systems are usually the way formal verification is done, I am not denying that. The problem I am claiming is not the tools of formal verification but the domain one has to model and capture.

What I was trying to show is the nightmare of modeling accurately and in a way that matters lots of operating system primitives and further reasoning about when the assumptions have been violated by the program set up.

Indeed this does not effect the stuff outside the interaction of the program with the operating system and my point when it came to modern programming primitives is that not all data structures or methods of programming lend themselves to being able to write formal proofs about them.

You might be able to have a type BigInterfaceTypeThatDoesLotsOfThings but this might be impossible to formalize in any meaningful way and AI is not the panacea and this stuff is not easy and require hard thinking.

Additionally a lot of formally verified programs are done via constructive proofs where you build up the correct program from primitives and extract the program and prove it that way. This requires generally requires a type system that has a complete inference algorithm like ML type-systems and are much harder to work with and you have to set up knowing you want to do this sort of thing.

Capability systems indeed make a lot of sense for operating system level constraints and I agree this is a sensible direction and combined with type systems you can build some very powerful permission systems. But these do not solve formally verifying lots of behaviors and good capability system in my opinion (and something I worked on previous as a side project that is completely unfinished and abandoned.) need mathematical primitives and structures that lend themselves to proofs and being restricted by capabilities.

Note capabilities become a program themselves and types can become programs, so getting your types right has the same issues in the first place when talking about this level of formalization.


Not only can you switch easily, they all seem to do the same thing, fail in the same ways, make the same bad linguistic choices, write the same code... I think at the end of the day, LLMs are a commodity. They're like a grease, or oil. They're trying all the tricks with branding cute names, version numbers, benchmarks, thinking levels, but it all just seems the same stuff to me: I put in tokens, I get more tokens back, and if I didn't know ahead I wouldn't be able to distinguish if they'd come from Anthropic or OpenAI or wherever. Maybe some people can but I can't.

Not speculating about Tim Peters and why he died, but careers in CS promote a sedentary lifestyle, that might have something to do with it. At the same time, Alan Kay is 86!

These discussions always go the same way:

"We shouldn't be funding all this frivolous stuff!"

"Okay, how would you figure out what to fund?"

"We could assemble a panel of experts in the field, who have proven their merit by doing impactful research, and have them decide ultimately based on the merits of each proposal, the people who proposed it, the institutions backing them, the likelihood of success, whether they're worthy of funding. We should also make them explain as part of the proposal exactly how this will benefit not just their research area but also the broader community and world. Finally, we should meticulously track how they spend their money to make sure there's no funny business, and only give continued support to researchers who have proven they spend it wisely and their research has impact."

(If people don't understand, this is currently how it's done)


> We could assemble a panel of experts in the field, who have proven their merit by doing impactful research, and have them decide ultimately based on the merits of each proposal, the people who proposed it, the institutions backing them, the likelihood of success, whether they're worthy of funding. We should also make them explain as part of the proposal exactly how this will benefit not just their research area but also the broader community and world. Finally, we should meticulously track how they spend their money to make sure there's no funny business, and only give continued support to researchers who have proven they spend it wisely and their research"

This is not accurate. There is the first gatekeeper which is program managers and junior program managers which have to gloss over your project and pick you. They (especially the juniors) don't necessarily have a the sophistication to judge your project and so are easily swayed by hot topics or reaching claims of national security/health applications.

Grants get awarded to people in the habit of telling little white lies about what their research is good for, then its people who are awarded grants that get promoted, and then those are the people who are reviewing projects, so being dishonest becomes cultural. Disastrous for science.


How would you change the system so it doesn't create a culture of dishonesty and decisions aren't swayed by hot topics?

> couldn't explain to the public why half of these experiments were worth a single tax dollar.

The funny thing about that is the controversial “impact statements” that are part of grant funding are supposed to be the part where you explain how your research matters to the broad community. But it’s become so political the only thing you can feel safe writing is “we will help people learn about science” because “we’ll start an after school program for disadvantaged youth to learn about science” is too DEI for some folks.


That's not exactly true when you look at the history of how fascism works. Actually, capitulation is futile, as there is no bottom to what the autocrat will demand. If you give in, all it shows is you are an easy mark. Resistance is the only winning move.

Timothy Snyder put out a great book about this called "On Freedom", which explains most of the power an autocrat has is given freely by people who anticipate what he would want and just offer it without being asked, precisely because they have a notion that resistance is futile. Tech leaders were right there at the inauguration, they didn't even try! So the role these tech guys played early on was teaching Trump the extend of his power.

If you look at how Trump handled schools and law firms, you see an excellent case of capitulation vs. resistance and how that played out. Take the situation with Columbia compared to Harvard. With Columbia, they capitulated to Trump illegally withheld $400m and agreed to all sorts of demands, only to be met with more demands. Harvard faced similar pressure, but decided to fight it and won in court.

In the law world, Trump made demands to several high-profile law firms. Paul Weiss, Skadden, and Kirkland & Ellis all capitulated to the demands, other firms saw these law firms capitulate and just offered up settlements themselves. 11 law firms entered into agreements with the administration, and they didn't even have to ask all of them! Meanwhile, every firm that challenged demands made by the Trump admin, including Perkins Coie and Jenner & Block, won in court.

Tech giants didn't capitulate to the Trump administration because they felt they had no other choice, they did it because they felt it would be best for their bottom line. They have no principles, no backbone, and are opportunists who will say anything if a dollar is on the line.


Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: