IIRC, for "normal" bit widths the codegen basically uses the next larger machine type and preserves zero bits on the high end. An i3 is an i8 with five MSB zeroes. It's UB to fill those with non-zero values. For larger bit widths, like u729, you concatenate many large machine types, the compiler generates instructions in an unrolled loop, and the LLVM optimization pass usually doesn't clean that up (though, now that integers are apparently not using the LLVM u729 implementation, perhaps there are some more optimization opportunities).
They're situationally useful, especially when performance isn't an enormous concern. That u729 example above came from a variant sudoku solver I wrote to aid developing new puzzles (easy to check the rough magnitude of the solution space for whatever idea I was mulling over and examine how restricted the board actually was -- just an intermediate step in puzzle design). It's not optimal (hard on the icache, can be hard on registers, other issues abound), but it's dead simple to use, and the assembly isn't terrible, beating all the normal solvers I saw floating around. It's a nice point on the laziness/correctness/good-enough-perf pareto curve.
Another comment mentioned this, but they're great in packed structs for representing weird numeric entities (I think I have a logarithmic number system floating around which does that).
One thing the language does quite a lot is use them to guard against certain classes of human error at compile time. It doesn't perfectly make impossible actions unrepresentable, but shoving a full u32 into a shift argument usually doesn't make sense, so the types are constrained to be smaller.
> internet training data is not where frontier capabilities come from
We 100% would not be at the current progress without it, though. And it's not like they only train on this once. They keep training on all the internet data PLUS the private data. Private data only (probably) wouldn't work, as learning the base regularities of language takes a lot of weights.
Definitely. Despite the wording of my question, which I intended to be rhetorical, I understand why people are doing it. While I think that is one explanation, I think it's mostly hype or optimism. In other words, not so much grappling, but rather hoping and wanting AI to do anything and everything.
I just don't personally get why people want that. This endemic force feeding of AI down everyone's throats at every corner is just amplifying people's worst tendencies: greed, laziness, apathy, cognitive disengagement, etc. It's especially apparent when people suggest AI automating activities that have always largely been about personal gratification, growth, and/or expressions of thoughts to share with fellow humans.
I won't deny the benefits that some people get from LLMs (primarily experienced senior developers), but I feel those benefits are far and away outweighed by costs that it's already having on everyone else, both individually and societally.
I think GP found it useful for this particular use case, as did I and my coworker who is even more enthusiastic about it (he calls it compressed human culture).
And no, Google search (or Wikipedia) won't come close to capacity of LLMs to find similar things.
How does something with 232 comments and 207 points over just 2 hours get pushed to the 3rd page in hacker news? I’m just really curious how it works, like why would something with so much engagement be push down so quickly?
> Its like they've been chomping at the bit for decades to get rid of those pesky humans and are so hyped up over it they can't see clearly anymore.
This is precisely it. Here's my analysis:
AGI is a savior figure for the capitalist class. A tech version of the Second Coming, delivering them from the pesky demands of workers, like a living wage or (gasp!) sick leave.
That's why they're all so obsessed with it, it has religious-ideological component to them. When you hear them talk about AGI, there's always this weird eschatological vibe with it.
Unfortunately, they're blinded by their beliefs and can't think things through even one step further. Even if their cyberjesus comes down to them through the machine and replaces all workers, who's gonna buy all their stuff then?
All they're doing in their capitalist zealotry is ringing in the end of capitalism.
I thought Apple usually locked in contracts with TSMC and Samsung for years in the future? They should be best positioned to weather this storm. If they are getting buffeted enough to raise prices by this much, things are going to be dire for smaller manufacturers.
Or, this could just be a convenient excuse to get even more margin.
It was also designed by European engineers, not in Michigan. Not saying that's the reason the Focus is more reliable than a Taurus but they didn't follow the "typical" Ford design process at the time for that vehicle. For what it is worth I owned a 1992 Taurus and it left me stranded more times than I can count. Just some of the issues I had were a water pump that exploded and a seized A/C compressor.
porsche is part of volkswagen, so it's not that surprising that they're decently reliable. i probably see 10 porsches for every ferrari, lamborghini, etc that i see, and i think a large part of that is reliability - even absurdly rich people don't want to deal with an unreliable car when there is a more reliable alternative.
Look at the AWS Prime ssds available, and it's a massive list of strange drives you've never heard of, with very few reviews available, almost all using YMTC. The prices aren't fantastic, but given that five sixths of the drive market is straight up gone, I think we need to start reviewing and hoping for the best here, fast. I really hope endurance is indeed as rated, that we aren't about to all get burned incredibly badly for using YMTC chips.
CXMT is indeed starting to get some ram out there. But I am pretty skeptical it's going to make much of a dent. They're currently single digits % of the world ram production. They need to scale a lot to make any dent, and as soon as they do, it feels like there's plenty of people ready to snatch up those supplies.
We need massive massive massive growth in availability and there's no sign that current scale up plans will be at all adequate.
Current accelerators (TPUs, various onchip NPUs) are something close to this. Systolic array is the estabilished computer architecture term for flowing data from computation to computation without the overhead of a register file or von Neumann bottleneck.
hey guys! I've been working with a friend on this OS for a while now. I was curious if anyone could leave honest feedback about the direction of the OS.
- How do you transform and enrich the data? How does your pipeline look?
- What are your key challenges?
- Which tools do you use? What is your 'stack'? (Stanze, wordfreq, Whisper, wn, ...)
Background: I am currently building a multi-lang vocabulary hub for language learning. The goal is to match core words/lemmas to their senses/concepts, and then be able to generate multi-language flash cards.
I am still stuck on the sense alignment and fingerprinting (example: should 'to shop', 'einkaufen', ' alışveriş yapmak' and 'go shopping' point to the same concept of 'shop'?), but in a later stage I want to allow user-submission and data enrichment for IPA, pictograms [1] and audio.
Use-case (the dream): I come back from language class, I input new vocab and I output new Anki cards that work across all my fluent languages.
Currently, I mostly find myself knee-deep in problems of linguistics, NLP, Python and getting an LLM to do exactly what I want. At the same time it is a super fun project, and really makes me feel the joy of programming again. LLMs are magic, time just flies by, and all the random projects I always wanted to do suddenly materialize.
For coding, I mostly use free Gemini and some deepseek-v4-flash via openrouter to keep a tight oversight and understand the problem space. Maybe this slows me down, but agentic code jsut does not align with me. Overall, I haven't spent more than 2 € in total.
So far, surprisingly, the biggest problem is the lack of high-quality, free input data (example: English has the Oxford 5000 words as core vocabulary, but it is difficult to find the same for e.g. Turkish).
2nd place is the lack of high-quality synsets/wordnets (cross-language is mostly incomplete), and the 3rd place is getting LLMs to reliable play to their strength (on paper, a LLM is the perfect tool to provide multi-lang sense equivalents)
I plan to do a full writeup sometimes, but first I need it to work :)
