> I love how everyone is acting now that like in good ol' days(2022) every website was made by a hand with virtually no bugs but a lot of love.
Please point out where I said that. It’s not that "in the good old days" websites were perfect, it’s that every single website coded in 15 min with a LLM like this one tries to look great but is full of random bugs that a normal human would have seen if they had done this by hand.
I hope it does get fixed, but Apple actually has a bad track-record with another deprecated API, AppleAL (OpenAL). They broke it in a very interesting way in Tahoe where mono-channel audio on Aarch64 (not through Rosetta) did not play and it was broken through all the betas. I believe it actually may still be broken today. The workarounds were either mono-to-faux-stereo channel duplication or switch to openal-soft.
OpenGL is an even bigger deal so I don't see a world where it doesn't get fixed, but I also thought exactly the same way with the previous issue. It seems that Apple does not have any older video games in their testing apparatus.
OpenAL was terrible on Android as well, but audio on Android was terrible at the time anyway. Game developers had issues like explosion sounds playing half a second after the visual effect. Google made excuses for years saying it’s not their fault somehow, while it was fine on iOS, before they eventually fixed a scheduling mistake they introduced.
I think at work I am just barely in the category of I've removed more code than I have added. Over many, many bugs and long meetings I've sort of arrived at the conclusion that in most cases, the best code (excluding comments) is the code that doesn't have to exist [0]. It occurred to me at work today while refactoring some change that even in this LLM-driven world, I can still optimize for that for some additional challenge and satisfaction.
The root of this challenge comes from more text aka more code ~= more tokens, so these firms have little incentive to optimize for that mindset. Sure, they might have more optimized individual responses - but the end result is still giving people what they want: more features, more documentation, nice, long, verbose and well-formatted Jira comments, of sometimes questionable but sometimes undeniably impressive quality. The band marches on, and when your codebase outgrows the constraints of the tried-and-true tools, you can just invent your own. With documentation and an MCP or two, or ten -- to also pull into your context window.
[0] I work on an operating system, so a suggested retort is to say this is me pushing responsibility and logic to the people higher up on the stack ;-)
> The root of this challenge comes from more text aka more code ~= more tokens, so these firms have little incentive to optimize for that mindset.
I'm not sure if this is true. The code is such a small part of the total output. Thinking harder, like what would be needed for architecturing better code, would use more token.
With AMD, most GPUs including the integrated ones will just default to the `amdgpu` module which is maintained by AMD themselves. You pretty much have the best possible experience on Linux right now.
I can't speak from personal experience since my only laptop is an Intel Framework 13 with no dGPU, but there'll be two other main cases both involving Nvidia that tend to bite people.
1. "Gaming" laptops usually with integrated Intel GPU + Nvidia discrete GPU. Not sure what advancements have been made in power management, there. I know there's some incantations one can put in .desktop files to prefer the discrete GPU for like, games, but not much more I can speak to.
2. General fun involving the fact that Nvidia's "gold-standard" drivers are not mainlined in the kernel, but maybe the fact that they are "open source"* means at some point there'll be parity with them. I think that is the goal of Red Hat's Nova driver in development.
* these days, the GPUs got complicated enough they have management processors on them that can actually run most of the complicated algorithms via a firmware blob. So the current open-source drivers basically bridge the gap between the kernel's various subsystems and that little processor to work the magic.
On another note, Intel Xe/i915 have been pretty good, I've used both for integrated GPUs and an Arc Alchemist-generation discrete card of theirs. The framework 13 12th gen intel GPU struggled at the start with a lot of hangs but within a few months of launch, some smart cookies at Intel managed to fix various bugs and it was smooth sailing from there, with the exception of thermals just being in a rough spot on that generation.
Eventually it will, but it is console exclusive for some indeterminate amount of time. Google seems to indicate that this was 19 months for GTA5. Not entirely sure why, the tinfoil hat theory seems to be it's to encourage people to buy it a second time to play multiplayer with their PC friends. Alternatively, it could just be it's easier to get some fixed targets out the door than worry about broad optimizations.
A bit disappointing either way though as someone whose only console is a Nintendo Switch 2.
Considering the low end will be the Xbox Series S, that puts it not too far out from the Switch 2.
However, the CPU would be way behind the and storage on Switch is somewhat limited. I don't think people would want 180GB chewing up the 232GB of available storage on the system.
It would be in the impossible port territory but it seems like it could be JUST within grasp. This is all dependent on how well the Series S performance is. If it stinks on there, I doubt they will ever bother with a S2 release.
This feels like a nitpick but it's important to mention anyway
> Mullvad Browser uses them [ the DoH servers ] by default when you're not on Mullvad VPN, preventing your ISP from seeing the domains you visit.
This is a half-truth until Encrypted Client Hello (ECH) is in-use for most of your traffic. Sure, you won't have clear-text DNS floating out there in the series of tubes. However, without ECH the hostname you're connecting to is exposed in the SNI field during TLS handshakes.
I imagine most sites we are browsing will eventually be able to leverage ECH, so this should become less of an issue over time, but in the event you frequent any self-hosted operations (e.g. no Cloudflare or similar), encrypted client hello could still reveal identifying information since it requires still exposing an "outer SNI" that a valid certificate needs to be presented for. You only get to easily hide in with the masses when you use a big infrastructure provider as a consequence, since the common outer SNI of cloudflare-ech.com doesn't really reveal anything different than an IP <-> ASN database would. But if the outer SNI is "dunder-cat-enterprises.com", then all I've protected you from is your ISP snooping on the subdomain or pinpointing exactly which one of my domains you might be talking to.
Even ECH isn't that helpful. ISP still sees the IP addresses you connect to. Even when non-dedicated IPs are used, I'd be surprised if a quite basic traffic analysis (say, bytes transferred on first visit) wouldn't identify the domain.
VPN providers at the tier of Mullvad should be precise about this stuff -- I think it's more than just a nitpick, considering the audience. oh god did I just use an emdash.
Note that you need some flavor of secure DNS to enforce ECH. The protocol is designed to be downgradable.
Yes, but the article (not to call you out - I just think it's a very important point!) points out that this type of throttling would not be effective:
> Suddenly, the crawlers were coming from millions of random residential or mobile IPs, all pretending to be random modern browsers. An IP like that would make 4-5 requests and then never show up in the logs again. There was no point in banning them, because by the time you figured out that they were bots, they were already done with you. You just needlessly ballooned your firewall ruleset by adding IPs that would never be back.
Without something like cookies (which are almost certainly tossed after the IP is rotated) or some other persistent identifier, you are stuck have to apply mitigations that scale with the load you're encountering, which means longer challenges for everyone or degraded functionality, like removing some of the fancier cgit features.
I've had a fair bit of success with increasing the bot mitigation based on a global rate limit. During periods of high request rates, I throw progressively more hurdles at the bots, and during periods of low request rates I disable them all.
I testified against the equivalent of this bill in my state. One of the things I mentioned is because of a non-trivial monetary fine per infraction [1], as someone who would potentially need to implement this, I would have no idea how to reliably differentiate a computer that was simply out-of-date/implementation has bugs from willful non-compliance vs Open Source [2].
It just felt like the bill had the goals it had when it was created, and the broader cloud of "gee, if people implement this a certain way, it could have unintended consequences" was completely ignored. But don't worry, one of the co-sponsors proclaimed. near the end of the hearing, that they had a Masters Degree in Computer Science and worked on operating systems in their career, so they made sure to let us know that we were over-reacting.
The other thing that really pissed me off was it was rumored my states bill was going to get an open source exemption. However, they waited until the end of the hearing to introduce all of the amendments, including the Open Source exemption. The proposed amendments were not publicly visible on the bill page or the page about the committee meeting for that day. This ended up being an excellent strategy to sway other committee members [3], since they could just hand wave the bulk of us as "concerns resolved". It was quite illuminating to also see media lobbyists come in and verbatim just state "hey did you get our proposed amendments?" and then without much reading of them at all, they were ratified during that session. Lesson learned: the moment there was a rumor of amendment, I should have made a considerable effort to get my hands on that text ahead of time since for the implementation concerns I raised, I was still unhappy. I'm not quite sure though how much of a difference it would have made though, since at least 30 other people fell into the same trap.
Anyhow, hopefully my long-winded Sunday morning post was useful to someone in the future either when dealing with compliance, a version of the bill in their own states, or the federal government's attempts to do a similar thing. I confess that I have been lazy and not remotely done any due diligence on that federal bill (https://www.congress.gov/bill/119th-congress/house-bill/8250... / HR-8250: Parent's Decide Act). I encourage others to be better than me and contact their representatives, assuming it isn't already on the fast track to becoming law.
[1] at least, for smaller businesses -- big tech won't care! I think it was something like $6,000/pop, which is chump change for big tech and will be negotiated during settlement talks.
[2] Okay, you can check the user agents. But who wants to need to maintain or pull in a list of enforceable user agents! What happens if someone is spoofing the UA and suddenly they end up in my list of "must check" UAs (or inversely, !(not must check)). How long does one reasonably wait for the API call to time out? What about running the app on Wine? And most importantly, what if someone that doesn't care about Linux ("okay claude make my website legal no mistakes") is in charge of implementing this logic, or the library that people will end up using for compliance.
[3] Assuming they cared. The vote was largely amongst partisan lines, some people had clearly looked checked out the whole time, despite the abnormally high numbers of their constituents being there that day.
In fact, it seems to be the case for the "json" and "llms.txt" buttons, too.
reply