qalmakka

I don't really agree with this. Having strong types and guardrails like in Haskell, Rust or OCaml helps _massively_ with LLMs because they get very clear messages back, and can express their logic using semantic types they can easily follow. Conversely, I've noticed that AIs (just like humans) tend to kind of lose the thread with very dynamic code

> What if I told you that Docker Desktop has always used microVMs?

I'd say that I don't think it's that relevant? Hopefully everyone was aware that you required a VM to run Linux software on Windows and Mac, and then who would really care about what VM docker used to run development containers? The more expendable the better. Hopefully nobody is using a Windows PC with docker desktop in a production scenario - but given how utterly incompetent some people I met was, I'd not be too surprised from that

Yes, but at least it was a real thing with a serious proposal behind. Even if Carbon ships, what value would it give in 2029 or whenever it is in a world where writing rust or heck even C++ is now way easier and foolproof (as long as you know where to guide an LLM)

In my experience LLMs are way better when they "know" a language "instinctively". It's just that unless your language is very niche, the corpus is usually good enough. I tried using Claude to write my own personal language a while ago (I wrote a toy compiler decades ago) and it struggled a bit, because you could see in it's reasoning it had to "repeat" the syntax equivalence to itself while it read the code. It didn't just "know" it could use a given construct to do something; conversely Astra, when carefully instructed to do so, can plop down esoteric template code that works the first time, because it just "knows" it's the right stuff to write

Carbon was clearly DOA the moment it was announced, IMHO. It looked cool but it served none but Google, and now with LLMs you have a massive incentive not to use a niche or new language due to how better LLMs get the bigger the corpus is

The only somewhat realistic proposal in this space is Herb Sutter's cpp2, which is arguably a massive improvement and I'm puzzled why nobody in the standard thought to give it a spin, there's just to much cruft they'll never be able to get rid of unless they make an alternate yet backward compatible syntax with C++ that changes the defaults from "random 80s nonsense" to something better

yeah it does feel like a Lisp that people happen to write in an imperative fashion most of the time

Tcl is way too smart of a language for its own good. I love how everything is a string or can be hacked on a string, so you can basically do levels of metaprogramming that reach unheard of levels

> secretly in a cash crunch

What's secret in the fact that literally they've never had a single positive quarter yet, and we're a rack of GPUs being repossesses away from going from a bull to a bear market? Their entire business is based on getting financing from VC, and they've been keeping the lights on by pushing any narrative they could to sustain the hype. Regulatory capture isn't really a conspiracy theory, it's the very exact thing multiple companies in their position have done in the past to avoid a market readjustment they knew they'd lose

I think that the economic side of things of how those businesses are run, and the fact AI has potential to be an incredibly powerful tool, are orthogonal concepts. I don't see how the fact an LLM can be used, autonomously or directly, to perform hacking or penetrate computer systems, has to do with the fact that there's clearly an interest from those labs in being strictly regulated while dictating the rules they should respect in the first place. I suspect such regulations will be surprisingly difficult to implement for anyone but the existing US-based labs, and of course it will be super surprising when it turns out that since the federal government stopped everybody from making new frontier models they suddenly got super profitable.

> What would you say when their models cause new incidents like hugging face on

What would you say if Caterpillar builds a bulldozer with a faulty breaking system that kills people? I'd hold them responsible because they failed to oversee the development of their product like every responsible company should do. If they're so much in a hurry to ship that they can't even test what they're building properly, they must be held responsible for that - and I say this as one of their customers and users, not as a random Joe that dislikes LLMs. LLMs are useful but it's undeniable that these labs are clearly desperately trying to keep the lights on no matter the consequences

Both OpenAI and Anthropic are locked into an arms race funded on shaky hype-based debt. They have to constantly push the envelope at the frontier or risk seeing their funding pulled, probably causing a global financial crash in the process which would kill them.

Clearly they've realised that people are increasingly fine with non-frontier models, including Chinese ones. They can't consolidate and just deliver a product to profit because they're constantly being caught up by the Chinese at prices they will never be able to match.

I suspect their main goal for the last year and a half or so was to pull a GM - let the share of the AI economy grow, then have the government bail them out when the debt becomes unsustainable. Clearly Trump would have never let the bubble explode before the midterms, right?

Well it's now clear that the US government has way too much debt right now for 2008-style bailouts, and the market doesnt have the money lying around to buy their overinflated stocks at an IPO. The only avenue realistically left to them is to cry wolf so much to get governments to ban all non-approved models, stop new developments and give them breathing room to consolidate to get their current crop of models profitable at higher prices.

Unfortunately for them I suspect Mr President has way too much dementia and way to little business acumen left to realise the sham going on, so I think they'll keep doing charades for a while until they'll suddenly stop

Both OpenAI and Anthropic are locked into an arms race funded on shaky hype-based debt. They have to constantly push the envelope at the frontier or risk seeing their funding pulled, probably causing a global financial crash in the process which would kill them.

Clearly they've realised that people are increasingly fine with non-frontier models, including Chinese ones. They can't consolidate and just deliver a product to profit because they're constantly being caught up by the Chinese at prices they will never be able to match.

I suspect their main goal for the last year and a half or so was to pull a GM - let the share of the AI economy grow, then have the government bail them out when the debt becomes unsustainable. Clearly Trump would have never let the bubble explode before the midterms, right?

Well it's now clear that the US government has way too much debt right now for 2008-style bailouts, and the market doesnt have the money lying around to buy their overinflated stocks at an IPO. The only avenue realistically left to them is to cry wolf so much to get governments to ban all non-approved models, stop new developments and give them breathing room to consolidate to get their current crop of models profitable at higher prices.

Unfortunately for them I suspect Mr President has way too much dementia and way to little business acumen left to realise the sham going on, so I think they'll keep doing charades for a while until they'll suddenly stop

The major advantage of CLI is IMHO that you can use a CLI yourself too. It's a universal tool.

One thing I hate is when people write an MCP server for something useful and just that. Like, why don't you tell the same Claude or Codex instance you wrote to also expose the same functionalities as a CLI? The problem is that it's full of technical people with zero CLI knowledge nowadays (or, rather, for the last 20 years or so), and it kinda sucks in general, because lots of stuff only comes in GUIs that are not composable, not scriptable and sure as heck full of stupid bugs and limitations due to how harder it is to get GUIs right compared to popping out a "dowhateverctl" kind of tool

I am no fan of Iran but the white people in question literally went there to wage war with the so called has-been-empire and got their butts spanked, despite EVERYBODY warning them for years this would have been an unwinnable war

This is basically self inflicted, as usual messing up in the Middle East always ends up with consumer paying the price for their ruler's stupidity

"Oh no, my drinking water tastes like crap after I pooped in the well"

> CUDA for AMD on Windows

This gives me strong "Bash on Ubuntu on Windows" kind of vibes

Codegen was still done via LLVM. LLVM supports the MSVC _runtime_, here they're talking about using their (arguably worse) backend directly to generate code and do optimisations

... but why? The MSVC backend has been falling far behind LLVM with every release. I understand they want uniformity but IMHO either they catch up or they switch fully to LLVM, if they can

For me it definitely was

libuv does all sorts of shenanigans, like using the undocumented fields in STARTUPINFOW the CRT uses to pass file descriptors to processes inheriting handles so that you can open whatever file descriptor you need in your children processes like in POSIX

But Microsoft kind of expects this crap and will support it until the end of time; Apple can't give a crap until and will readily break programs even when they use legal APIs

I don't know. I know that the old interface was less pleasant than it could be, but breaking every single program ever written in Zig to change it for a super unwieldy API felt to me more like the authors being unnecessarily fastidious than striving for "no hidden control flow". Every single language has a shitty API or two - it's impossible to have none, and to be fair the new one is way too clunky anyway.

Like, what does "avoiding hidden control flow" even mean? What value there is in having no convenient facility to do buffered I/O? Hidden buffered output wasn't really a problem even 40 years ago, it's one of those things everybody needs and those who don't are well aware of how to use the primitives and do it properly.

But let's be clear, I have nothing against Zig, I quite like it from a language nerd standpoint and I think it's very neat - I just don't understand what is the intended target to be honest. If I'm not looking for a safe language, C is fine, and for all its warts and bullshit C++ works with basically everything under the sun. If I want safety, there is Rust which basically already enforces the same patterns you need to follow to get safe C/C++ programs anyway - in my experience, 99% of the time someone whines that the borrow checker is stopping them from doing something, it's because they are making a massive logic mistake and they didn't notice there's some hidden unsoundness (like pointers with unclear lifetimes, etc).

One thing XML is very good for is semantic documents adhering to a well defined schema - and it's one of the few commonly used text-based data storage formats that do it well (with XSD, DTDs are pretty terrible). Unfortunately the only library that currently implements validation is libxml2 - expat is way nicer, but it never supported it. I don't understand why most XML libraries do not implement validation - XML sucks as a plain data storage format anyway.

It's mostly useful to allow sandboxed AI apps to access external functionality. If an LLM has any kind of access to a CLI then it stops making sense, because LLMs are very good at CLI and almost always they use less tokens doing that. Plus, you human can easily use a cli tool, while mcp is just for agents.

This is clearly not how it's going though. Any advancement from any lab has been quickly (< 6 months) matched up by basically everybody else. Even Grok nowadays is decent, and that's something. When something like you've described actually happened historically you generally had quite fast a clear frontrunner and a bunch of copycats that failed miserably; in 2026 we are very far from that. we are heading face-first into towards a pricing war because all models are easily interchangeable nowadays - AI is turning into a commodity more or less

I never truly understood what the intended business model around LLMs was. Get them widespread through cheap pricing and then jacking it up? Being the only ones that had a viable product so to get the ability to extract as much value as you want from AI?

I don't understand how a product that:

- is interfaced with and is deeply linked to natural language, so everything you produce (sessions, history, etc) is in Markdown and you can literally install a second model and tell it "hey import all of Claude's memory into yours" and that's it

- is based on well understood technology, the real constraints are how much money you put into training the models, but the theory has all been developed in the open

- clearly has a threshold where it quickly commoditises and turns from "I want the best" to "hey the best is a bit too expensive. The second best is half the price and works close enough".

was ever supposed to be a money printing machine. The fact something is extremely useful doesn't imply it's extremely profitable.

IMHO we're clearly speedrunning the process of turning AI into a commodity. Dario Amodei knows pretty well that when or if Anthropic cuts people off Fable, the vast majority of them will definitely not pay for it because Opus 4.8 is good enough for almost everybody that _knows_ what they're doing, and so are basically half of the most recent models. If I already have good baking skills I don't become more productive with an automatic bread machine, I just need a better dough mixer and oven

If those buses use DEF, have a anti particulate filter and are well maintained, their emissions should in theory be extremely low, with no heavy compounds. I hope local municipal authorities don't remove filters from their buses. Even then, most if not the new buses in my town are either electric or CNG/electric hybrids

It is a serious crime. Diesel exhaust is s IARC type 1 carcinogen. Somehow people go nuts when companies pollute water but not when they pollute the air they breathe

To be fair is quite hard to actually understand what this White House is saying most of the time

It has been clear to every single human being reading it for hundreds of years. It was never up to debate until some person of this administration decided to go and question the meaning of "jurisdiction thereof". Also saying that a foreign-born person isn't subject to the jurisdiction of the country they find themselves in opens a massive can of worms - like, if the State doesn't have jurisdiction over foreign nationals, does it imply it's not legally allowed to arrest them for instance? Two can play the same game and find infinite loopholes even in the clearest of texts.

This goes beyond the value of citizenship by birth, which I'm neither in favour nor against (personally I think that just sanguinis is nonsensical, but so is to automatically give citizenship even to accidental passer-bys), it's all about whether the law still carries any "evident" meaning or whether it can be spun around depending on political necessity, which is bad

This is kind of crazy. The text of the amendment is as literal as it gets. If this had failed it would have basically meant emptying Congress of all of its powers, because now the executive can just pick whatever interpretation they deem fit to their goals and run with it.

The rule of courts of law is to interpret the law, not to pick new creative meanings out of them. That's the role of the legislative power - otherwise what's stopping a court to reinterpret the meaning of any word in any legal text and allow the executive to rule by decree

To be fair a lot of it had also to do with the sheer immense amount of vast, mostly unused ,fertile land available in north America. I sincerely doubt the American experiment would have worked this well if they had rowdy neighbours and infighting due to resource constraints. For almost 200 years the solution to most things in the USA was to get a chunk of either their people or immigrant to move to the neck of the woods to find fortune