• 0 Posts
  • 302 Comments
Joined 2 years ago
cake
Cake day: June 18th, 2023

help-circle



  • Maps was them underestimating how much work it is to create good map material. The functionality was fine from the beginning if I recall correctly.

    Apple Intelligence is them panicking because the rest of the industry started putting more ML/AI features on their smartphones and they weren’t just late to the party, they apparently barely even started working on it.

    They put their own twist on it with “Private Cloud Compute” (make of that what you will, the theoretical tech behind it is an interesting read though), and they also want to process many features entirely on-device (again in the name of privacy, but to be fair Gemini Nano also runs on-device).

    Then they realized that running somewhat complex ML models on device requires memory and that’s where they always cheaped out on their products, so when they announced Apple Intelligence at WWDC in summer (with new iPhones only being announced in September) they had ONE iPhone model that could even run Apple Intelligence: the iPhone 15 Pro. So you could’ve bought an iPhone 15 (non “Pro”) the day before and every single feature they announced aside from tinted home screen icons or whatever wouldn’t work on your brand new device.

    They announced a whole bunch of features, the biggest one probably being a new Siri that has a “deep understanding” of the appointments, email, photos, messages etc. on device. This has now been delayed to iOS 19 or whatever.

    The other (smaller) features have been drip-fed over the iOS 18.x releases. Also, Apple Intelligence works in the EU starting with the iOS 18.4 beta. They said that it was delayed because of EU regulations but I think it was just a convenient alibi and it just wasn’t ready earlier on their part.

    I live in the EU, own a 16 Pro (so the “latest and greatest” iPhone) and installed the 18.4 beta to check Apple Intelligence out. And let me tell you “beta” is an understatement. I enabled Apple Intelligence and it said it needed to download models and that the phone should be connected to a charger. I did that and monitored network traffic in my router. Once major network activity stopped I checked but nothing. Waited for another 1-2 hours, nothing. Disconnected from the charger and then several hours later my phone shows a notification that Apple Intelligence was now ready.

    So, what’s there? Hard to say exactly but it summarizes emails but only some of them and I can’t make out a pattern. The quality of the summaries has been okay for me, but often times not much more useful than the subject line.

    You can hold down the camera button to open something resembling Google Lens, but the functionality seems to be limited to “send what I see to Google Images” or “ask ChatGPT about this image”.

    I’m not sure if notification summaries are in Apple Intelligence already because I never got any summaries (I also think it’s pretty useless as most notifications are already a summary of something).

    Then there’s an image generator (“Playground”) but it’s very limited. It is kind of neat to quickly put a portrait of yourself in a couple of different settings though.

    There’s also an emoji generator called “Genmoji” and sure it kind of produces okay results, but my iPhone tends to completely shit itself when I use it, slowing to a crawl and killing background apps presumably because it’s running out of memory. They (pretend to) want to do the most ML stuff locally out of everyone but but the least amount of RAM in their devices (8 GB in the 16 Pro, 16 GB in the Pixel 9 Pro).

    I switched to iPhone (from an Android device) in 2016 with the original iPhone SE (with A9 SoC), had an 8, 11 Pro, 13 Pro and now 16 Pro. They’ve all been a good to great experience including the latest software features, but iOS 18 on the 16 Pro isn’t it. Even if I turn off Apple Intelligence completely, iOS 18 is pretty messy: the icon tinting sometimes gets stuck so when switching light/dark mode some icons stay in the other mode, only fixed by restarting the device; Igot more random resprings than with any other iOS version; the front camera sometimes takes 10+ seconds to start working and then has a 1 second shutter lag from time to time, etc.



  • Mumble, or maybe TeamSpeak 6 (they skipped 4, had 5 in beta, which now is 6 in beta, oh well).

    Depends on what you want. We’ve been using a TeamSpeak (3) server I’m hosting for years, it works as well as ever (they added a couple of QoL features to the TeamSpeak 3 client during the pandemic as well).

    TeamSpeak 6 supports persistent chat via the Matrix protocol and you can register to any server and use that to login to any server using federation (as it uses Matrix under the hood). They now added screen sharing so you got the features covered that most users would want. They unfortunately didn’t release self-hostable TS6 server yet (but they say they’re working on it) so you can either use an experimental TS5 server (uses Matrix but doesn’t support screen sharing) or TS3 server, which doesn’t support any of the new stuff. The TS6 client is backwards compatible though.

    I just don’t think they actually know where they want to go with it yet. They seem to be advertising the whole decentralized thing as that’s clearly a differentiating factor from Discord, but on the other hand they didn’t exactly prioritize putting out easy-to-setup server software yet. The TS6 client pretty much fully supports TS3 servers including administration, but as far as I know TS6 servers are quite a bit different. There’s also “communities” that work with TS6 servers in some way. So it’s all a bit of a messy mix between legacy support and their attempt at creating a decentralized Discord.

    I hope they get it together and release TS6 server software, find a good way to monetize their efforts and get people to use it.

    Some people will say that you could just use Matrix directly instead, but if they manage to make TS6 easy to use and understand, allow easy creation of a server (as a service) and also allow full-featured self-hosting it could turn out well. Plus they have the brand recognition, at least with folks that aren’t that young anymore. This might help with adoption. Sure, it’s proprietary still, but it’s decentralized and uses open protocols (Matrix). You can apparently already join TeamSpeak community chats from your own Matrix server, so they aren’t artificially blocking “vanilla” Matrix servers from federating.


  • It would probably take a lot of information to its grave, but the more known “servers” would probably get crawled by archive teams.

    Also - assuming Discord wouldn’t be replaced by something equally closed off from easy public access - all new information would be easier to access.

    When Discord started, they marketed it primarily as a voice chat software for gaming. I remember them marketing it as “superior audio quality to TeamSpeak” or similar wording (which by the way wasn’t the case). It obviously has chat, video chat and screen sharing conveniently built in which TeamSpeak is only starting to add now in 2025 with the TS6 beta (they seem kind of lost atm).

    I always preferred the decentralized nature of TeamSpeak and Mumble though and at least from my own experience, TS tends to work better with fewer connection issues and better autogain and voice leveling.

    I don’t like the fact that most people happily gave up decentralized voice chat for a centralized alternative and we still use TeamSpeak in most of my circles to this day.




  • Fabric with some performance-enhancing mods is a great choice as well, yes! I’ve been wanting to test it on my server for a while now, just haven’t got around to it yet.

    Paper changes some of the more quirky vanilla redstone behavior, although - again - it’s very configurable so some of that original behavior can be restored.

    I’d mostly base it on which plugin/mod ecosystem you prefer/require.


  • World simulation (ticks) is single-threaded, but things like world generation are multithreaded. I’d recommend Paper as server software as it’s more performant out of the box (vs. vanilla) and configurable (ex. how many threads world generation is allowed to use).

    If you host multiple worlds I recommend spinning up a Paper instance for each world separately and connect them with Velocity.

    Ryzen 7000 should have better single-threaded performance than your i5-9500 but as it’s a VM ymmv depending on whether Sparked Host overprovisions their machines.










  • Speculative execution seems to be the source of a lot of security flaws in many different CPUs. CPU manufacturers seem to be so focused on winning the performance race that security aware architecture design takes the backseat.

    Also, it’s more and more clear that it’s a bad idea that websites can just execute arbitrary code. The JS APIs are way too powerful and complex nowadays. Maybe websites and apps should’ve stayed separate concepts instead of merging into “web apps”.

    I also wonder if it’d be possible to design a CPU so vulnerabilities like these are fixable instead of just “mitigable”. Similar to how you can reprogram an FPGA. I have no clue how chip design works though, but please feel free to reply if you know more about this.