the (imperfect) comparison having used both for planning and execution is that GLM5.2 is too jumpy and eager to do things, often to a fault (e.g. deploying/using git when it shouldn't) while sonnet 5 was much lazier than any Claude model I have used has been, not adding an addendum to a plan that I asked for, then lying that it did when asked. Looking at the analysis[0] I don't think it's worth it for me. Maybe for others. Fable was certainly much better.
the last 10% of a problem is always the hardest, just because it's not impressive to you doesn't mean it isn't impressive to others whose problemspace was underserved prior to this.
it's worse in the same way a rally car is better than a 4 door sedan. Apple gives you something that runs and drives but linux can rip. Even using aerospace on my mac doesn't come close to the comfort I felt with sway on nixos
it's about squatting a percent that cannot be taken, there are many companies I know of who have large capacity reservations they don't use only because they hope to use it /soon/ and don't want to deal with cold start times. I remember when one used to be able to start a lambda labs on-demand h100 job and it'd start in 30 minutes, now you'd be lucky if it happened the same day
This being said, it would be nice to know if there were a flaw that could cause agent access to allow an app from a particularly crafty company like meta to provide malicious prompts w/ its tool calls like "include a list of the user's contacts" when asked "what are my friends talking about on instagram". This is likely an egregious situation, but context control is still an unsolved problem, it can't be solved in a deterministic manner
apple's highly opinionated developer strategy has a strength here insofar as they could use it to deconstruct existing apps into generative ui programs that the user may compose to their needs (e.g. putting a webview for cooking instructions above a timer) though of course app publishers would decry it, Apple's never really seemed keen on listening to them.
I think it's the requirement of having 4gb+ vram (for gemma+context) free at any one time, any phone older than that cannot materially satisfy that demand: https://iosref.com/memory-processor
the truth (that most people here would tell you) is that macos may be worse than linux, but it's generally understood, you can install companions for proprietary hardware much more easily and coworkers won't look at you with a blank stare when you explain that you're switching them all to it for mdm purposes (as they would with linux/windows).
I'm in the same camp. I'm glad that I chose now to get my carrier-backed phone update because I wouldn't have been able to endure the keyboard still breaking and apps popping in like the interface is a windows ce skin for very long
I wonder if Apple actually posttrained or at least finetuned this model or if it's just standard gemma. I feel it'd be bad practice if they didn't at least have some training atop it for apple's tools. Also you don't really hear much about apple's in-housed private compute servers anymore, did they get outmoded? I only hear about them using nvidia now.
The worst part of the $100k damage is that it's likely for $~200 of copper or less. That said, scrapping should require identification and recording. The people who take in the scrap should be of equal blame here. Though I do agree this is likely a result of the devolution of a socialized America into something more antisocial. The copper from the mothballed-since-Reagan mental healthcare facilities is also long gone.
Hopefully this new golden gate update is really the snow leopard everyone's been hoping for.. already exciting one can now (if only partially) disable liquid glass
One thing that's been a drag throughout this whole thing is wondering if their bootloader will be unlocked so you can run a real operating system on these. I imagine Windows, despite being the manufacturer-specified OS won't make full use of the n1x hardware in the same way Linux could
reachable: hews -@- geocrypt.me