Apple is on the side of privacy if it serves their marketing. Which is why they would rather build and normalize CLIENT SIDE CONTENT SCANNING so they can continue to market iCloud as "secure and private".
If Apple's interests sometimes align with ours then great. I'll take it. But don't attribute to this ~5 trillion dollar company some kind of altruism.
> Apple has abruptly withdrawn its lawsuit against NSO Group, citing increased risk that the legal battle might unintentionally reveal sensitive vulnerability data and difficulties in acquiring essential information from the spyware vendor.
> In a court filing Friday, Apple said continuing the lawsuit now poses “too significant a risk” of exposing the anti-exploitation and threat intelligence efforts needed to fend off the very adversaries involved in the legal dispute.
> “When it filed this lawsuit nearly three years ago, Apple recognized that it would involve sharing information with third parties. However, developments since then have reshaped the risk landscape associated with sharing such information,” the Cupertino device maker said.
I mean, iOS has never been open source so Apple has always practiced security through obscurity. It doesn't seem unreasonable that they'd be concerned about this. Idk, seems like a nothingburger but maybe I'm wrong.
Dropping the case is great for Apple's obscurity, but terrible for enforcing the security of iOS users that are still vulnerable to Pegasus malware. Now NSO Group is not threatened or deterred, which is the worst of all worlds for iOS security.
This[1] study suggests image framerate does not significantly affect our ability to perform predicted moves accurately in time, above a certain minimum which is lower than 24 Hz.
Moving the mouse and then pressing a button, or pressing one key and then another, are both cases of predictive moves.
So it seems one shouldn't rely on framerate when talking about the limits of sensing input lag, at least in general.
Current top 5 played games on the Steam Deck are: Slay the Spire 2, The Binding of Isaac, Dave the Diver, Stardew Valley, and Baldur's Gate 3. BG3 is the only one you can consider AAA, but Larian is hardly an EA or Blizzard.
The big AAA studios recycle the same formulas and push graphical quality (mostly downstream of Unreal Engine improvements) because they are risk averse. Similar to big comic-book-hero films. It's not what consumers want and it's increasingly starting to show. AAA is struggling badly while indie has been on an absolute tear the past few years.
On top of that, the seemingly endless remakes... Some are praise worthy others not so much. I got tired of Resident Evil Remakes, they are just recycling the same formula over, and over, and over again... Innovation has left a long time a go. Other example is the recent Assassin's Creed Black Flag "Resync" (which is a remaster with some things on top of it), I was planing on buying it, but Jesus, Ubisoft couldn't let it pass: their launcher is still there, all DLC's together cost more than the base game (and why have DLC in the first place?), the menu is has an ad for Assassin's Creed Shadows which, etc.
I'm currently doing a course (I think it's equivalent to an "Associate degree" in USA) in Game Development. I feel demotivated to take part in this industry, and I think that, at least for me, it should be just on and off gig to do in my spare time, perhaps with other enthusiasts too, but I don't plan to pursue it.
I guess Valve would half-qualify as AAA? The top played true AAA game would be, imho "Grand Theft Auto V Enhanced", at the 18th place. Perhaps Ubisoft with a Assasin's creed title in the top sellers probably should be higher in the playing list.
But I think it's fair to say AAA doesn't hold the steam store anywhere near as tightly as it holds the consoles.
That was my experience a couple months ago, and until someone shows me real evidence of something valuable they've made with it, I'm not wasting my time on this stuff again.
GLM 5.1 was the model that made me feel like the Chinese models had truly caught up. I cancelled my Claude Max subscription and genuinely have not missed it at all.
Some people seem to agree and some don't, but I think that indicates we're just down to your specific domain and usage patterns rather than the SOTA models being objectively better like they clearly used to be.
Perhaps not even necessarily subjective, just performance is highly task-dependent and even variable within tasks. People get objectively different experiences, and assume one or another is better, but it's basically random.
Unless you're looking at something like a pass@100 benchmark, the benchmarks are confounded heavily by a likelihood of a "golden path" retrieval within their capabilities. This is on top of uncertainties like how well your task within a domain maps to the relevant test sets, as well as factors like context fullness and context complexity (heavy list of relevant complex instructions can weigh on capabilities in different ways than e.g. having a history where there's prior unrelated tasks still in context).
The best tests are your own custom personal-task-relevant standardized tests (which the best models can't saturate, so aiming for less than 70% pass rate in the best case).
All this is to say that most people are not doing the latter and their vibes are heavily confounded to the point of being mostly meaningless.
The pass@100 is such a weird critique angle that is surprisingly mainstream; guess what, no one cares if the correct answer is in the top 100, it needs to be the top 1. A model with a better answer in the top 1 is a better model, full stop.
This. Plus if you want to even attempt measuring real 'intelligence' you want to run a neuro-symbolic, de-lexicalized benchmark (e.g. DL-ReasonSuite, SoLT, GSM-Symbolic) - which none of the providers releasing new models showcase.
>just performance is highly task-dependent and even variable within tasks. People get objectively different experiences, and assume one or another is better, but it's basically random.
You are right that this is not exactly subjectivity, but I think for most people it feels like it. We don't have good benchmarks (imo), we read a lot about other people's experiences, and we have our own. I think certain models are going to be objectively better at certain tasks, it's just our ability to know which currently is impaired.
They might be converging somewhat. The ultimate limiting factor is training data. Eventually I think they will converge and then the competition will be on memory and compute efficiency, with the best being the smallest maximally capable model.
People judge models on their outputs, but how you like to prompt has a tremendous impact on those outputs and explains why people have wildly different experiences with the same model.
I had one occasion where GLM 5.1 did about 95% of the implementation that I needed but couldn't progress form there. And Codex (free quota) solved the remaining 5% on the spot. I'm super happy with both. I don't touch anything Anthropic with a 10 foot pole.
>GLM 5.1 was the model that made me feel like the Chinese models had truly caught up. I cancelled my Claude Max subscription and genuinely have not missed it at all.
GLM 5.1 is pretty good but there are some "buts".
They hiked the prices 2 times this year. I subscribed to the pro coding plan just before the last hike. At the start of the year, they had only 5 hours quota and no weekly quota. And I hit the weekly quota hard. I can't upgrade the subscription to get a higher weekly quota because they jacked up the prices a lot recently.
My $30 subscription costs now $72. Previously was $15.
Max was $49,then $80 and now $160.
The value in Claude Code is its harness. I've tried the desktop app and found it was absolutely terrible in comparison. Like, the very nature of it being a separate codebase is already enough to completely throw off its performance compared to the CLI. Nuts.
- Claude Code has repeatedly had enormous token wastage bugs. Its agent interactions are also inefficient. These are the cause of many of the reports of "single prompt blew through 5-hour quota" even though it's a reasonable prompt.
- It still lacks support for industry standards such as AGENTS.md
- Extremely limited customization
- Lots of bugs including often making it impossible to view pre-compaction messages inside Claude Code.
- Obvious one: can't easily switch between Claude and non-Claude models
- Resource usage
More than anything, I haven't found a single thing that Pi does worse. All of it is just straight up better or the same.
I've found GLM 5.1 to be extremely good, as long as you keep the context under 100k or so. But I do agree; the actual service is rough. It's often unusable during peak hours.
I imagine they're trying to slow down the acquisition of new customers because they're so overloaded. But yeah at double the price it doesn't seem worth it anymore.
Linux’s ecosystem has also improved significantly over the past two years, especially in China. Due to the influence of “Xinchuang” (that is, domestically produced Linux rebranded under another shell), many Chinese desktop applications have been reworked in the past couple of years, switching from Windows-specific tech stacks to cross-platform ones—mostly Electron, basically browser wrappers—and now support the Linux platform. The commonly used software is basically all there.
In addition, the development of LLMs has greatly lowered the barrier to using the Linux command line. Problems that used to take a full day to solve can now be handled easily by anyone who can write a prompt—just ask, copy, and paste. This has even made Windows’ command line unfriendly by comparison, despite its own major improvements in recent years, turning it into a significant drawback.
If Apple's interests sometimes align with ours then great. I'll take it. But don't attribute to this ~5 trillion dollar company some kind of altruism.
reply