1,764 karma · joined December 23, 2021
If you meant they’re now better at mimicking compilers, sure, but they’re only mimicks.
Compilers are mechanical and engineered to produce a correct output. A compiler emitting incorrect machine code is exceedingly rare, and considered a bug. They have heuristics and probabilities in them, but those are to pick between a set of known-good outputs.
An AI is a bag of weights outputting a probability of the most plausible token that follows [1]. It is inherently probabilistic in nature and its output is organic (by design, they’re designed to mimic human speech), as opposed to mechanical like a compiler.
A compiler follows hard rules. An AI does its best.
And to be fair, AIs are no better than human in this regard: humans are pretty bad at generating correct code without mechanical tools to keep them in line (compilers, linters, formatters). It’s not a wonder we use the same tools to keep LLM output in line as we do humans. (And, to be fair, LLMs are better than humans at oneshotting valid code).
[1]: to those that tell me this vision of an LLM is outdated: nope. The heavy lifting is done in the probability generation. Debates about understanding are not relevant here, and the net output of an LLM is a probability vector over raw tokens. This basic description can be contrasted to a compiler whose output is a glorified Jinja template.
Oh well, just like caps lock can be remapped, so can my keycaps be swapped (perks of blank keyboards I guess), though it'd be even harder to use a keyboard that's not mine I guess.
I somewhat get the argument, but if you're writing code in the HN textarea you're doing something wrong (for code where tab/space matters anyways). Like, any code editor will use the tab key properly.
Though, it sills maddens me there's no somewhat universal tab-entry in OSes like we have with enter (somewhat because there's a mix of shift+enter, alt+enter and cmd+enter). All of shift/alt/ctrl tab are usually also hijacked.
Yet the article doesn’t tackle at all the hard part: making alerts that are actually meaningful. They handwave it instead of giving actual advice. This post is a good intro, but I didn’t "walk away" with anything useful.
This is why, in this case, AI is important. Someone puts in an effort to write a short article (if a bit wordy) that can be used by e.g. beginners or managers? Good! I’m not the target audience. But if it’s the output of AI, what’s the intent?
However, to be fair, that's only at lower DPIs smudging's an issue. Anything retina-ish and integer scaling loses all meaning. I'm typing this on a 15" 4K laptop on 1.75 scaling with HN set to 120% zoom. At those DPIs, it does not matter at all. I adjust most websites zoom level because a lot of them think the content must breathe, while I think I should not fiddle with my scroll to read a paragraph.
- The frames in frames per second are dimensionless, thus the SI unit for FPS is frames/s. When the frames are periodic, such as monitor refresh rates are, the unit is Hz.
- Percentages are dimensionless quantities too, produced by divinding two quantities of the same dimension (ie unit). CPU%? That’s "busy second per second", which is dimensionless, and expressed as a percentage.
A dimensionless quantities don’t have any physical backing reality in terms of the, well, dimension in which you could measure it. Time, space, mass, etc.
Fun fact: angles are dimensionless! Both degrees and radians are just shorthands as divisions of the unit circle.
Given the current tech, I also doubt there will be practical uses and I hope we’ll see the opposite of what I wrote. But given the current industry, I fully trust them so somehow fill their hardware.
Market history shows us than when the cost of something goes down, we do more with the same amount, not the same thing with less. But I deeply hope to be wrong here and the memory market will relax.
Because the IP address changed, so classic routing still works. Their point is about identifying a session with something non-constant (the IP of the client), rather than a session token.
Instead of identifying the "TCP" socket with (src ip, src port, dst ip, dst port), they use (src uuid, dst uuid) which allows flows to keep working when you change IP addresses. Just like you can change networks and still have your browser still logged in to most websites.
The packets carrying those UUIDs still are regular old IP packets, UDP in the case of QUIC. Only the server needs to track anything, and only has to change the dst ip of outgoing packets.
As for flooding and DDoS, that’s what handshakes are for, and QUIC already does it (disclaimer: never dug deep in how QUIC works so I can’t explain the mechanism here).
That is the sad reality of the future of memory.
However when the infra is crumbling under requests (one could say decaying), then becquerel seems appropriate
Live migration boils down to copy memory over the network, stream the page faults till you converge enough, and resume execution on the other host. It’s not a hard problem but a precise and tedious one.
Pausing a game might involve a lot of GPU contexts to freeze, network resources to pause, storage streams to pause, input handling, sound, etc. Add to that physics engine that may be tied deeply in the system and you end up with a hard problem.
What a VM does is not the role of the hypervisor, thus it can apply its hammer that works in pretty much all cases, and VMs are pretty much all the same. On the other hand, all games are bespoke with custom plugins and custom integrations, which make them the opposite of "generic pause implementation".
Mailchimp is specifically made for mass email emission, for marketing a newsletter and whatnot. So yeah, a lot of people will consider them spammers.
In fact, in all cases, you don't know when the syscall actually starts execution even with regular calls. The only thing you're sure is the kernel "knows" about the syscall you want. However, you have absolutely no indication on whether it started to run or not.
The real question is: are the classical measures accurate? All we have is an upper bound on the time it took: I fired the write at t0 and finished reading the response at t1. This does not really change with io_uring. Batches will mostly change one fact: multiple measurements will share a t0, and possibly a t1 when multiple replies arrive at once.
Is it important? Yes and no. The most important thing in such benchmarks is for the added delay to be consistent between measurements, and when it starts to break down. So it's important if you're chasing every µs in the stack, but not if your goal is lowering the p99 which happens under heavy load. In this case, consistency between measurements is paramount in order to get histograms and such that make sense.
The state of Wayland compositors move fast, so support may be here. Last thing I’m waiting for is sway 1.12 that will bring HDR.
Incredibly, Linux has better support than windows for it on the desktop: DWM runs full blast, while sway supports VRR on the desktop. Windows will only enable it for games (and games that support it). Disclaimer: Wayland compositor required.
It’s not enabled by default on e.g. sway because on some GPU and monitor combos, it can make the display flicker. But if you can, give it a try!
My desktop runs Arch with Sway (so quite close), three monitors, and uses ~400MB ram after boot. Most of it are the framebuffers. All the rest is eaten by Firefox, rust-analyzer and qemu.
So basically, no use when you've not tasted 120+Hz displays. And don't because once you do, you won't go back.
On the capacitor though, the capacitor level is weird as you don't build the capacitor charge system with transistors. Though I definitely get that the simulation engine is for digital stuff, not analog :)
Also a general feedback on the time-based challenges: dial them back. A lot. Most of them are just not interesting and have zero learning value. In fact, the "DRAM refresh" one just made me quit the game (clicking on 8 rows to keep them fresh). Okay, 10s is enough, I got the point. No need to hold up for a whole minute. Kinda same for the hex one. However, some of them are good, and the UI for the binary ones is great, especially for the two's complement one!
Small nitpick on the UI: some blocks don't have their connections aligned with the grid, making the wiring OCD-incompatible. But that's minor. It's a shame since the wire routing algorithm works quite well overall, and I'm impressed an LLM could produce that good of an UI!
Otherwise, quite a fun little game, if slow paced when one already knows some bits of digital logic. Keep up!