Marvin

Rogue01 said:

Siri and improved Apple AI reminds me of AirPower.  Schiller - we can do this.  Schiller a year later - no we can't.  I am not holding my breath for Siri.  I rarely use it because it isn't very reliable, and the response is usually the same thing, let me see what I can find on the internet.  I can do that myself and get a better response.  Apple has had 14 years to fix Siri, and hasn't done it yet.

Local AI is too resource-intensive for mobile just now. The 7 billion parameter models use about 10GB RAM. Apple's local models have around 3 billion parameters:

https://machinelearning.apple.com/research/introducing-apple-foundation-models

GPT 4 is 1.7 trillion parameters. GPT 4o mini is 8 billion:

https://deepnewz.com/ai-modeling/microsoft-paper-reveals-gpt-4o-mini-size-8b-parameters-gpt-4-1-76t-claude-3-5-a05ca5f5

If they can get something like GPT 4o mini to run locally, that would be an improvement. iPhones would likely need 16GB RAM.

They want agent-like capability with Siri to be able to do tasks (1:44) where someone can ask the device to do everything:



It needs a few things: more RAM, better reliability from smaller models, good training data and process, good QA testing.

It feels like a good default setup would be normal Siri for local tasks but instead of searching the web when it has difficulty, it uses GPT 4o on the server and has local Siri interpret the reply.

comcastsucks said:

Clair Obscur: Expedition 33 or we riot!!

This runs ok on Crossover on Pro/Max chips:



It would be good if Crossover had an agreement with Valve for the Mac version of Steam to install and run these games more easily and the developer could see if it was worth doing a native port. It's on Unreal Engine so most of the game should be able to export to the Mac platform.

It makes sense that Apple would wait for a major OS to add a new MetalFX but they probably could have added it to the current OS and the games could launch earlier, now it will be September/October. They wouldn't want to launch Cyberpunk without frame-gen/interpolation. If they are using AMD's FSR 4, that should allow 3-4x framerate.

There's a way to patch the new GPTK into Crossover and it was tested here with frame-gen turned on with Cyberpunk:



They said it boosts FPS from 30-40 to 60-80 so maybe it's capped at 2x frames. Still a good improvement to the smoothness on lower-end Macs and the native build should run even better.

kkqd1337 said:

Smells like a paid for statistic 

They could just have used ratings from major review services for the TV shows.

Apple TV+
https://www.imdb.com/title/tt10986410/ (Ted Lasso - 8.8/10)
https://www.imdb.com/title/tt5875444/ (Slow Horses - 8.3/10)
https://www.imdb.com/title/tt14688458/ (Silo - 8.1/10)
https://www.imdb.com/title/tt11280740/ (Severance - 8.7/10)

Netflix
https://www.imdb.com/title/tt27995113/ (Black Doves - 7.2/10)
https://www.imdb.com/title/tt13016388/ (3 Body Problem - 7.5/10)
https://www.imdb.com/title/tt7221388/ (Cobra Kai - 8.4/10)
https://www.imdb.com/title/tt13649112/ (Baby Reindeer - 7.7/10)

The other services have much more content by a factor of 20x or more but there would be much less difference limiting it to shows rated 8/10 or above:

https://www.ampereanalysis.com/products/about/analytics-svod

Apple is trying to maintain a smaller library of highly rated shows. The services are low priced enough to get more than one service. Having both Netflix or Amazon Prime and Apple TV+ covers most things.

The lack of good quality shows is an industry-wide problem and not a new one. Cable was always described as having so many channels and nothing on:

https://www.youtube.com/watch?v=YAlDbP4tdqc

George Lucas commented on this too:



This affects the games industry as well as the movie industry. They start out with original ideas on a low budget but they are risky. A lot of studios passed on Star Wars:

https://www.vanityfair.com/hollywood/2015/12/star-wars-george-lucas-independent-film

Then some IP makes a lot of money and all they want to do is sequels/prequels/spin-offs. Sometimes sequels continue making money like the newer Star Wars movies, Top Gun and Mission Impossible but they are recycling the same concepts and it comes out less than the originals.

For decades people wanted a Back to the Future 4 and they fortunately declined to make it:

https://insidethemagic.net/2024/11/robert-zemeckis-breaks-silence-on-back-to-the-future-part-4-dr1/

It's like the old saying about customers wanting a faster horse. What people want is the originality that came from the old movies and if the world/characters are interesting enough, a sequel is justified.

Building a new, interesting IP is difficult, a lot of parts have to come together - the cast, story, cinematography. When it does all of them like Game of Thrones and the original blockbuster movies from the 80s/90s, it sets the bar for everyone else.

Foundation should have been Apple's Game of Thrones but the cast and scripting isn't quite there. Severance has received higher ratings even though it was a more risky theme.

AI tools will help out with this because they will be able to generate a photoreal animatic of every movie and TV show with the preferred cast and accurate voices and they can iterate on it in a matter of days. Then hand it to the production team to shoot.

There isn't an unlimited amount of original ideas though and this problem will get more difficult over time. This can be seen in Apple's recent movie The Fountain of Youth that is a play on the Indiana Jones theme:

https://thoughtcatalog.com/sergio-pereira/2025/06/apple-tvs-fountain-of-youth-was-a-blatant-indiana-jones-rip-off/

Most of the scenes in it felt like they were from a template. The car chase scene that follows the same structure, the choreographed hand-to-hand fight scenes with the comedic timing. Predictable and manufactured all the way through.

One thing they have a habit of doing with content today is making it feel unimportant like the cast and crew are in it for the money and it shows. The more that Apple's production studios get a feel for what works, they can direct their budget accordingly. More movies would be nice because TV shows are more forgettable/disposable.

s.metcalf said:

“If the product is free, you’re the product” seems to come to mind.

Why should they get my phone number?  It’s bad enough you have to give companies your phone number whenever you buy anything online, and I bet they get sold and abused all the time.  Scamming vulnerable people by text and phone is an increasingly serious problem.

There needs to be technology that can hide your phone number much like hide my email.

This app isn't collecting this data, this EULA is from their parent company. They likely just consolidated their EULA across all their companies.

They own companies like Zynga, their privacy policy links to the parent company's policy, which has the same text about what data they use:

https://zyngasupport.zendesk.com/hc/en-us/articles/360034758451-Personal-Data-Requests
https://www.take2games.com/privacy/en-US/

Some games require accounts similar to an Apple ID, especially online games and to verify the account, they use a phone number or email address.

People are right to be wary of data collection these days but it's rare that companies who aren't in the business of data monetization will abuse it. Sometimes companies have these agreements to protect them from legal challenges from the likes of the EU who will fine them for collection of data without consent so they explicitly ask for consent and then people get annoyed even though they signed up an account with an email address and likely used their phone number for two-factor authentication. They have to store these or you couldn't login to the account.

Apple has a similar privacy policy:

https://www.apple.com/legal/privacy/pdfs/apple-privacy-policy-en-ww.pdf

The reason this game is free and others are discounted is they are promoting the new Borderlands 4 game coming out as the article mentions.

charlesn said:

Personally, I give Cook the benefit of the doubt, though not without some serious questions that I'll mention in a minute. But he has been Apple CEO longer than Jobs 2.0, coming up on 14 years this October, without a hint of deceptive or dishonest behavior during that time. So to quickly throw Cook in with the Sculley/Spindler/Amelio era of failure and broken promises seems like a VERY cheap shot, and one not deserved by Cook. Sure, be angry... call this out as the major f-ck-up that it is and that it shouldn't have happened. That's all fair game. But don't jump to dishonest and disreputable when Cook has never been either in his 14 years of leadership. And aside from his past record, Tim just seems too damned cautious and conservative a leader to take a flyer on intentionally misleading the public.

Gruber has met the execs in person and can contact them directly. He could have reached out to them to ask how things were going and could even have done this in an interview this year.

Rather than find out what's going on, he decided to very publicly write a put down of their leadership and their integrity just because of a delay.

This is where the impersonal nature of online commentary gives people a different sense of what's appropriate. If Gruber had a coworker and he had some concerns about them, this would be like writing an offensive note and pinning on a noticeboard for everyone to see instead of talking to them directly.

Now the Apple execs will give their time to iJustine, MKBHD and Brian Tong.

charlesn said:

But yeah, I have questions. My first reaction to last year's WWDC timeline for "AI Siri" was: if Siri still sucks so badly after 14 YEARS, how is Apple going to achieve all this in less than a year? I know I have old posts from that time which said as much. And my second, more damning question Is: why did Apple wait until less than a month before it was supposed to deliver AI Siri to announce that it needed (A LOT) more time? The best it could do for an estimate was "within the coming year" which, technically speaking, means as late as the end of 2026, since that is "the coming year." Sorry, but even if Apple had, at the outset, wildly underestimated the challenges it faced, it would have known it was going to miss its delivery date months before it made the announcement. That's not something that creeps up by surprise. Did the Siri team keep insisting that it was going to get it done? We'll probably never know the real answer. 

There's been a lot of advances in AI in recent years and there are enough open source models and clients so it's much faster to build a high quality product. The DeepSeek model was made in a short timeframe:

https://www.unite.ai/how-deepseek-cracked-the-cost-barrier-with-5-6m/

The big AI models used a few million hours of GPU time. With 10,000 GPUs, this is under 6 months. Grok 3 used 200 million GPU hours and 100,000 GPUs, took about 3-4 months to train.

Apple can use their own chips. The others use GPUs like the Nvidia H100. Apple would need about 3-4x the amount of M3 Ultra vs H100 or a custom chip. They'd be able to train a large model in under 6 months but it would only work in the cloud.

They don't have to do it all themselves either. Siri could use other engines. I think the main challenge Apple is trying to tackle is having good, local models. Nobody else is doing that, they have everything going to the cloud. The issue is RAM so local models have to be smaller and less accurate. It would be best if they could stream the model from the SSD like in tiles/layers like a series of nested models than a single large model. Llama3-7b works pretty well but it needs about 20 different models with different specialties.

What Apple likely estimated was the timeframe to train a competitive product but because it's inherently unpredictable, they couldn't know the result of the training would be production-ready. The issue of hallucinations has only been studied recently, nobody knows exactly how the models work but they will figure it out eventually. Apple probably shouldn't worry too much about the output of the AI, they can always have a filter on sensitive topics and just return a standard reply like 'my responses are limited, would you like to use a cloud-based model'. People would be ok with that for now.