Hacker News

Top stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. Can gzip be a language model?(nathan.rs)
    93comments
  2. AI Has No Wisdom and Neither Will You(alexn.org)
    273comments
  3. MiMo v2.6(xiaomi.com)
    444comments
  4. Spymarks, Not Watermarks(brand.io)
    133comments
  5. AMD's random number generator can't generate a 0?(flatassembler.net)
    105comments
  6. I said no and Apple said yes(dbushell.com)
    371comments
  7. Attention is all you have(alicegg.tech)
    274comments
  8. Line Scan Photos from MUNI Heritage Weekend in San Francisco(lawrence.lu)
    6comments
  9. Transformers Explained Visually(poloclub.github.io)
    77comments
  10. What Sun got wrong(dtrace.org)
    361comments
  11. Verda (Finland) raises $189M in Series B(verda.com)
    21comments
  12. A font that reads what you wrote(rohanadwankar.github.io)
    25comments
  13. I don't want to read what you didn't write(colinbreck.com)
    345comments
  14. 9 Ads per Minute: FIFA Cup 26 – "the price of the beautiful game"(bristol.ac.uk)
    160comments
  15. MiMo-v2.6-Pro: Intelligence, Performance and Price Analysis(artificialanalysis.ai)
    45comments
  16. Engineering Memory: On learning to memorize first 100 digits of pi (2024)(gregorygundersen.com)
    24comments
  17. AI coding has made CI a bottleneck, so we reworked ours to keep up(linear.app)
    338comments
  18. Divide by depth for instant 3D(gabrieloc.com)
    34comments
  19. Looking forward to Git 2.56 – and 3.0(lwn.net)
    87comments
  20. NASA’s Mars Sample Return mission is dead(science.org)
    345comments
  21. What It's Like to Work in One of America's Data Centers(wsj.com)
    25comments
  22. Texas police department ordered to close for failing to provide public benefit(dallasnews.com)
    discuss
  23. A build graph that rolls dice(fzakaria.com)
    2comments
  24. World Wide Words(worldwidewords.org)
    3comments
  25. Claude Status – Elevated errors for multiple models(claude.com)
    99comments
  26. The Advisory Group on Mathematics and Artificial Intelligence(terrytao.wordpress.com)
    76comments
  27. HERMES radio enables voice and data communication over vast distances(ieee.org)
    65comments
  28. Python Workers are now generally available(cloudflare.com)
    39comments
  29. How do traffic signals work? (2019)(practical.engineering)
    70comments
  30. People Training OpenAI's AI Fired for Using AI to Train the AI(404media.co)
    discuss

macOS 27: Workaround to avoid downloading AI models and save storage

232 pointsby 23h agoreddit.com
119 comments
23h agoHN ↗

This type of stuff gives me PTSD flashbacks from trying timo de-crapify Windows.

22h agoHN ↗

I am tired to fight against every platform that I have to use

De macify Mac Degoogle Android Demacify ipad DeLLM chrome browsers local llm

Can we expect sane decisions from corpos about what we want rather than what latest thing they want to push on us?

22h agoHN ↗

Apple says their users make 2.5B Siri requests per day. Sounds like people want it?

22h agoHN ↗

Even though I do use it, a choice would be nice.

22h agoHN ↗

i wonder how many of them are like 100% of my requests?

- "set a timer for 14 minutes"

- "remind me to make a doctor's appointment tomorrow at noon"

- "turn off the bedroom lights"

that don't require any form of AI and have functioned perfectly well (aside from having a hard time when i mumble) for years on non-AI versions of Siri?

22h agoHN ↗

I suspect even some of the more "advanced" queries they demoed off, i.e., "what was that restaurant John and I talked about last week?" also not require LLMs whatsoever and could have been accomplished with a simple search.

21h agoHN ↗

I think you're going to need some kind of ML to turn that question into a structured search, and interpret the results.

21h agoHN ↗

Not necessarily. NLP is possible to do without any ML. See ELIZA [0] as an example, which used pattern matching and substitution to process natural language. A few simple rules can cover a majority of requests. It's still AI, but not ML.

Though I do agree that the ML-based NLP thst we have today is much more usable, albeit more compute-expensive.

[0]: https://en.wikipedia.org/wiki/ELIZA

22h agoHN ↗

This matches my use cases. Poor speech to text recognition so I don't trust siri to do anything non-trivial

20h agoHN ↗

Siri was always AI, but the popular connotation of "AI" changes all the time. It now means "LLMs built with transformers."

22h agoHN ↗

Do they have a breakdown for platforms? iOS vs. macOS, I'd bet most of it is iOS.

I don't have Siri enabled, I've never felt the need for it on desktop.

22h agoHN ↗

How many apple devices exist globally?

Does everyone use siri? How about keeping it opt-in?

21h agoHN ↗

I dont even have Siri enabled, but even less so on macOS.

21h agoHN ↗

If you have a car you are forced to because you can’t use Apple Car Play if you don’t enable Siri.

I also don’t use but I have to have it enabled to be able to use Google Maps, and Siri occasionally mistakes a conversation for being invited to listen and speak.

18h agoHN ↗

Curious. I have Siri turned off (just checked again) and I use Apple CarPlay daily. Maybe an EU thing.

16h agoHN ↗

you can’t use Apple Car Play if you don’t enable Siri.

It works for me, but I'm using wired carplay on an iphone 11 w/ios27. (In the US, Swapped to Siri Classic, then disabled it.)

Looking around online it seems doing it wired can bypass the Siri requirement. Which may work in my case because in the Ford headunit I can force wireless projection to be off and blocked (including turning off the wifi for the headunit). So some cars might be forcing wireless Carplay and triggering the Siri requirement.

21h agoHN ↗

2.5B Siri requests per day

that's like one request per day per user lolol

21h agoHN ↗

I get forced to use it While using CarPlay. I don't particularly want it, though, considering it still regularly fails at basic tasks like adding the currently playing song to a playlist.

20h agoHN ↗

The issue it's they make it very hard to remove it to clear the space used by it, i think you got to boot into recovery mode and do some risky crap, my macbook is almost out of space just from development cache..

20h agoHN ↗

Never met anyone who used siri out in public, to be honest.

17h agoHN ↗

Timers. Easily >50% of these requests are people setting timers.

21h agoHN ↗

One arguably don’t even need KDE any more; the coding models will throw together a perfectly adequate desktop environment.

21h agoHN ↗

Why should one use something adequate when something amazing exists?

21h agoHN ↗

Not sure what the point of this would be? Devalued street creds for running something nobody else does? Who cares?

19h agoHN ↗

Well, it's literally the reason Omarchy exists so some people like it...

13h agoHN ↗

If you’re gonna use KDE, just go with Windows. Same UI paradigm and, surprisingly, Windows looks better (that’s how bad KDE looks).

8h agoHN ↗

"look, MacOS is following the same trend as windows as to become measurably worse over time, with bloat, slop, user hostile/dark pattern, removing freedoms configuration and escape hatches"

"KDE is immune to those by design"

"Then use Windows"

I'm surely convinced!

21h agoHN ↗

no. buy a linux compatible laptop and deal with the UX downgrade.

20h agoHN ↗

What downgrade? A typical Gnome desktop is far more coherent than Windows or even macOS in some regards these days.

15h agoHN ↗

Yes, and you can do much better than GNOME, too.

19h agoHN ↗

There are definitely some cool features missing, but I'd take KDE over any other window manager without a second thought if it were possible.

12h agoHN ↗

I’ve found the downgrade surprisingly minimal on plasma 6 - and it’s far more customizable.

Thanks to marginal gains on MacOs (nee OS X) the Linux desktop gets closer more quickly than Apple can pull away.

12h agoHN ↗

Yes I use cosmic DE on Fedora. I didn't find it a downgrade at all! Despite not running an M chip it is butter smooth

22h agoHN ↗

Yeah, I’ll wait to update my computers until Apple gives us an actual choice.

21h agoHN ↗

Time to permanently offboard from Apple ecosystem then

21h agoHN ↗

Perhaps eventually. I know I’m in a “closed ecosystem,” and that removes a fair bit of choice in a variety of places. But Apple is known to walk bad decisions back at times. This update is an example, with liquid glass. Another recent one was the iCloud private email planned domain change being scrapped.

20h agoHN ↗

Liquid Glass was rushed out because their flagship WWDC feature, Apple Intelligence, wasn’t ready as expected. They knew they didn’t finish refining Glass as they had intended to. They had unfinished work with Glass that got pushed to this year due to the last minute change in direction. The refinements we got weren’t only the product of feedback and they already intended to revisit the design.

This situation is a bit different where they’ve taken several years delay now to refine their AI strategy before releasing it.

If they let users permanently opt out of their AI rollouts this early on, their stock will plummet. They’re going to focus on making it better to drive adoption and value instead of letting people remove themselves from exposure to what they have in the pipeline.

19h agoHN ↗

I mean they added a slider to make Liquid Glass slightly less bad, but the “good” end is just the old “reduce transparency” option. They’ve got a lot of progress to make, if they are ever going to give us back the pre-liquid-glass UI density.

21h agoHN ↗

macOS 27 is way faster and more stable than macOS 26 though, so unless you are on an ancient version it doesn't make sense to not upgrade.

21h agoHN ↗

I've hit some really weird window/workspace UI bugs so far in MacOS27. It's not no-brainer update yet, IMO.

19h agoHN ↗

It keeps putting my left mounted dock on the wrong monitor for some reason. never happened before.

15h agoHN ↗

That's what I use on my M1 Macbook Air too, works fine.

19h agoHN ↗

On one of my machines, I don’t know if I could even make it fit. So it’s a no-brainer to not update on that one, if they are going to force an extra 14GB download on me.

16h agoHN ↗

Being short on disk space is a pretty good reason to not upgrade.

I would rather keep my personal picture library on my local disk (with backups, of course) than gigabytes and gigabytes of Apple models.

They were the ones who thought that 256Gb was a reasonable number. And it actually did work pretty well, until they wanted to fill it with models.

12h agoHN ↗

Calling something cutting edge one year ago “ancient” is absolutely wild. To me that word means, like, pre Apple Silicon.

21h agoHN ↗

why is a local model any different than any other file they want to install with their os?

21h agoHN ↗

Well because it could be an extra ~20GB which is not nothing if you have a 256GB disk.

20h agoHN ↗

because some people don't want AI and would rather even the files were not present in their computer.

16h agoHN ↗

I do want (some, carefully curated as to certain usecases) AI. But I don't want to waste my diskspace for this AI.

20h agoHN ↗

It's 14GB of something I don't want, need, nor have any use for, on a machine that is notorious for making storage upgrades difficult.

19h agoHN ↗

By that logic why not fill the hard drive with OS data? This is a company mind you that still sells some laptops with a base configuration of 256GB, so 10-20GB isn't an insignificant amount to many users especially when it's taken by a feature they may not even use.

19h agoHN ↗

Friendly reminder:

Many apps (like dictation apps) ship with their own models totally separate from Apple Intelligence.

For example, Co-Typist [1] is very useful, but it requires a model installed with it, ranging from Gemma 3 1B at 0.8 GB to Gemma 4 26B A4B Pro at 15.7 GB. It recommended Gemma E2B (3.2 GB) for my 16 GB, M1 Pro MacBook Pro.

It works really well; the autocompletion is so good, it knows what I'm going to type before I do!

Here's the thing: you may not be aware of the amount of storage being used by models in various 3rd-party apps.

The open-source app What The Model [2] scans your machine and reports all the models it finds and gives you the total amount of storage being used.

[1]: https://cotypist.app

[2]: https://github.com/powtac/wtm

20h agoHN ↗

I still don't see a good reason to upgrade from Sequoia.

19h agoHN ↗

I was skipping Tahoe too after seeing all those posts about the differently rounded corners.

This time I took the plunge and upgraded, and I have to admit the whole system, especially Safari is much more snappy than before.

22h agoHN ↗

Off topic. Since old.reddit is now login walled - is there an alternative frontend?

6m agoHN ↗

Is there a version of this that uses more than 10% of my horizontal screen space? I have a 40in monitor, I don't need the website compressed to the aspect ratio of a phone.

20h agoHN ↗

Use "Old Reedit Redirect" for Firefox - it still lets you in without a login.

20h agoHN ↗

That's not a reliable source or reference. Read the disclaimer at the bottom.

20h agoHN ↗

Yes, I'd sanity-checked the answer on my machine, which is 15.5 GB on iOS 27.2: https://imgur.com/a/XVT8rme

Not all of the space used by Apple Intelligence is downloaded model size, which explains the reasonable difference.

20h agoHN ↗

Your reality is yours, not mine. https://imgur.com/a/3viouLd

I'm not upset, just pointing out TweakTown gave Apple's AI misleading information about Apple's products. You might have 30GB of wasted space on your 128GB drive due to Apple Intelligence features on your Mac (like me).

19h agoHN ↗

That's not a reliable source or reference. Read the disclaimer at the bottom.

The "Always verify important details. Siri is an Al that may make mistakes." is the new version of "your mileage may vary".

Also, I get a similar message when using Gemini, OpenAI and Claude.

I own a "Standard iPhone" and an M1 Mac; these numbers track.

13h agoHN ↗

Keep tilting at windmills. Wikipedia was also not a source

21h agoHN ↗

Apple intelligence is listed as 29 GB in the Storage settings. Which should be the amount of space you get back.

MacOS 27 is also diffrent in that it does still downloads the data for EU users and removes or hides options in menues to dissable functionality. This means users lose the said space when upgrading to 27.

21h agoHN ↗

Apple Intelligence is available on macOS in the EU. It's only disabled on iOS/iPadOS.

21h agoHN ↗

without a reddit account:

... can you also force Siri off by using Screen Time controls? I noticed that in the Screen Time menus, "Content and Privacy" restrictions can be toggled on. From there seems to be an additional "Siri" menu that lets me change the allowed Siri version from Siri Classic to Siri AI as well as options to turn off the global writing tools/assistance (which turns off the annoying Siri bubble on every textbox now).

I suspect forcing Siri off via those controls would work easier than using Low Data Mode or trying to force Siri to disable via language selection.

21h agoHN ↗

I'll say they definitely need to let people disable/offload the model, but if you haven't tried the new Siri, it's been great. Obviously it's not a state of the art LLM, but it's good enough, and the fact that it can search and take actions on your local data makes it incredibly useful. I definitely use it more on my iPhone where it's slower to get things done the traditional way, but I use it all the time now.

"What was the quote for my insurance claim?" - it'll look through your emails and pull the number with the source email so you can jump right to it.

The other day I asked "when did I arrive and leave Mexico City on my trip at the end of 2024?" and it nailed it based off my flight receipt.

Since learning Spanish I get a lot more Spanish stuff in my feeds, and if I e.g. don't understand a meme I'll just ask "explain the joke here" and it reads the screen and explains it.

You can have an article open in Safari and say "remind me to read this tonight" and it'll make a reminder with the link. Or you can modify reminders - "move my reminder to get groceries to tomorrow at the same time". Yesterday I told it to make me a reminder and it did, but I worded it in a clunky way so I followed it with "reword that to sound more natural" and it did it.

We've been trained for a long time to think Siri can't do anything but I urge you to try it once you're on the new one, it's quite good. And I'm sure I've barely touched all the things it can do. David Pogue published a pretty thorough test other day:

https://pogueman.substack.com/p/125-tests-of-the-new-ai-siri

19h agoHN ↗

A couple of things I've done with Apple Intelligence on the Mac:

* When I need a prescription refilled, the quickest way is to message my doctor using the patient portal. I created a Shortcut that lets me pick among 5 medications; Apple Intelligence then generates a message with the request to the clipboard which I paste into the message field and send.

* I have a Shortcut that creates a morning summary: Apple Intelligence generates the weather report. I also get what’s on today's calendar and any current or past-due Reminders.

Just scratching the surface here; will flesh these out more later and create some new ones now that I'm running the release version of Golden Gate and not the beta.

18h agoHN ↗

Not sure if I want Siri to know my prescription medication.

17h agoHN ↗

Not sure if I want Siri to know my prescription medication.

Using Shortcuts, a user can use the Cloud model via Private Cloud Compute, the Cloud Pro with "world knowledge, and the on-device model. It occurred to me I probably don't need the Cloud; I switched to the on-device model, which worked fine.

There's no way for Siri or anybody else to know anything about a transaction using Private Cloud Compute:

    We designed Private Cloud Compute to make several guarantees about
    the way it handles user data [1]:
    
    A user’s device sends data to PCC for the sole, exclusive purpose of
    fulfilling the user’s inference request. PCC uses that data only to
    perform the operations requested by the user. User data stays on the
    PCC nodes that are processing the request only until the response is
    returned. PCC deletes the user’s data after fulfilling the request,
    and no user data is retained in any form after the response is
    returned. User data is never available to Apple — even to staff with
    administrative access to the production service or hardware.

[1]: https://security.apple.com/blog/private-cloud-compute/

16h agoHN ↗

I am sorry no amount of assurances from any company is going to make me reveal my medical information to products from any tech company.

11h agoHN ↗

I am sorry no amount of assurances from any company is going to make me reveal my medical information to products from any tech company.

Unlike your insurance provider or credit card company, which keep your prescription details, Apple Intelligence deletes the data once you receive a response—a fact Apple could verify if necessary.

It appears I did not need the cloud; my prescription information stayed on my iPhone to create the request, thanks to the Apple Foundation models running locally.

9h agoHN ↗

Luckily I live in a country with universal healthcare and our pharmacies don’t transmit prescription medication info to credit cards. They just show some prescription medication was purchased

7h agoHN ↗

what a cute anecdote. much success to your assurances

18h agoHN ↗

The Apple Intelligence Shortcuts creation is great as well. I never had a desire to learn their visual programming language so I mostly ignored Shortcuts. Now it's one-shot some really handy utilities for me.

18h agoHN ↗

Is it able to work with non default apps like Gmail?

18h agoHN ↗

Not the Gmail app, but a Gmail account, yes. You have to at least have the default Mail app installed and signed in for it to be able to index the emails.

18h agoHN ↗

Siri AI is very capable, but *I* should have control over when and whether *I* decide to enable or completely delete it.

I wasn't expecting Apple to force-feed the AI into my mouth with this new Siri release, but turning "Apple Intelligence" into "Siri AI" and removing the opt-out toggle is gross.

(You can't disable all of other fka Apple Intelligence stuff without disabling Siri AI, and you can't disable Siri AI without switching your Siri language to an unsupported region and turning a bunch of stuff off in Content Privacy and Restrictions. On Watch, Siri is front and center of your Suggested Apps carousel --- which they removed Recent Apps to create, an extremely useful feature --- unless you change YOUR ENTIRE WATCH'S REGION to some other unsupported country. None of this is okay.)

Anecdotally, I can do all of the stuff you've listed can be done without Siri AI or "Classic" Siri and get correct answers almost 100% of the time.

18h agoHN ↗

Like all macos features you used to like/depend on...

"privacy is a right"

becomes

privacy is a "defaults write <obscure-key>"

16h agoHN ↗

As a side note, the Apple Watch changes have completely made Camera Remote unusable for me, a feature I loved.

I don't know why, but I usually can only get Camera Remote to work when I close all of the recent apps on my Apple Watch, even when my watch and phone were the newest ones Apple released. Probably a bug with memory management they're never going to fix. Now, that's not possible, even when I reboot my watch :(

13h agoHN ↗

I agree that you should have the ability to remove AI features, but that isn't the direction Apple has followed for the past decade or so. There's a lot of stuff in macOS that one should be able to remove, or at least disable, if one isn't interested in living within the Apple ecosystem.

12h agoHN ↗

Anecdotally, I can do all of the stuff you've listed can be done without Siri AI or "Classic" Siri and get correct answers almost 100% of the time

You mean the way that takes 20x as long?

7h agoHN ↗

Better for it to take 20x as long and be right than to be 20x faster but be wrong while looking right...

18m agoHN ↗

The one that is almost just as fast workflow wise, far more capable, has more features, is developed at a faster pace, and is not forced on the user screwing their privacy.

16h agoHN ↗

Our experiences diverge surprisingly far. The new Siri was worse for everything I use it for before I turned it off.

The new Siri can not reliably return the weather (it seems it's prompting on the literal text of the town and asks for a clarification sometimes about whether I am in the United States or United Kingdom), and when I ask "what is the air quality index", the new Siri would sometimes return the definition of AQI. When it returns a value, new Siri would hallucinate an incorrect and fixed value. In the cases where it does work, it's very much slower than the classic "rainbow orb" Siri.

10m agoHN ↗

I've never even tried to ask Siri about the weather because I have a widget for that to the left of my homescreen. And it's on my watch.

15h agoHN ↗

"What was the quote for my insurance claim?" - it'll look through your emails and pull the number with the source email so you can jump right to it.

The other day I asked "when did I arrive and leave Mexico City on my trip at the end of 2024?" and it nailed it based off my flight receipt.

Where you see convenience, I see a security nightmare.

14h agoHN ↗

Why, factoring in the design of Apple Intelligence?

10h agoHN ↗

Doing inference locally is a significant improvement, but other issues remain, like prompt injection and data exfiltration.

Especially if the whole point is to have it search through a giant pile of arbitrary un-vetted stuff, like all the emails you ever received from anybody.

13h agoHN ↗

He says, browsing the Internet, probably with a Google account.

10h agoHN ↗

(Whispering) multiple things can, in fact, be bad

8h agoHN ↗

(speaking normally, as this is the internet and not an orchestra) No one said, in fact, that they couldn’t

9h agoHN ↗

I get that these are things you might do regularly and feel more empowered to use Siri for them... but they're rather banal and could be done with literally a one or two step process previously, especially if you're already using Apple's first-party apps which seem to be a requirement here. In both email cases, Spotlight handles searches like that just fine, and moving a calendar event only requires clicking and dragging.

I'm here wondering why so much extra space is warranted, or how to take it back, not so much how I can find a document with specific text in it or use Siri to do something else that was already very easy.

One task I'm wondering about is for example, asking Siri (on the mac) to analyze a huge dump of arbitrary PDFs that may or may not have non-OCRed text in them, and efficiently reorg and rename them based on what the documents should be. This also seems like a mundane task, but requires a little more than just a standard search and would allow someone to take their entire paper archive of private business documents and create a digital archive without too much risk or work. This is something that all local models I've tried so far really struggle with.

11m agoHN ↗

I don't know which Spotlight search you've been using but it certainly has not been consistent and reliable in years.

And I don't find manually searching for emails to find something like flight dates that easy. I've flown to/from Mexico city 6 times in the past couple years, and the airline tends to send like 12 emails pertaining to every trip I book, so asking Siri is much faster.

Not to mention these are tasks I'll be able to do in winter while wearing gloves.

8h agoHN ↗

I'll say they definitely need to let people disable/offload the model

If you have used mac/iOS (esp. Mac; well, because on iOS you can't do anything other than what's on screen essentially) for a long time you might realise Apple, at this point, is simply not capable of even developing those features that will allow users to do such things. And you don't even have to know Swift/objC for this sad realisation. I am pretty sure someone would have mentioned and they'd have realised it meant moving around a million hardcoded modules and frameworks and turning on/off a billion flags (a lot of them potentially global). They are that incompetent when it comes to software.

20h agoHN ↗

They need an option to completely disable and delete Apple Intelligence.

A 256GB Mac Mini is effectively unable to use macOS 27 due to this bloated nonsense.

19h agoHN ↗

Yeah I use my 256gb Mac Mini for iOS dev, which has always had enough space for an iPad and iPhone simulator with plenty to spare, but this eats way into that.

19h agoHN ↗

There's a distinction worth drawing here: "on-device" and "local" are not the same thing, and this thread is really about the gap between them. A model Apple pushes to your disk that you can't inspect, swap, or remove, that runs when the OS decides, is on-device in the literal sense and gives you none of the control that makes local AI worth wanting. Local, in the sense people actually mean it, is that you chose the weights, you chose when it runs, and you can delete it. Forcing a 14-29GB model you can't uninstall is the inverse of that. The storage complaint is real, but the deeper one is that "we put AI on your device" quietly became "we put our AI on your device."

18h agoHN ↗

Tangentially related, I've always been frustrated by Apple's refusal to allow sorting photos/videos by file size. It feels like a blatant push to force users towards paying for their cloud storage.

16h agoHN ↗

Not ideal, but their disk space recovery thing in settings can do I believe.

17h agoHN ↗

If Siri and Apple Intelligence is disabled in the settings, does it still do the download?

17h agoHN ↗

Worth pointing out that the Foundational Models can be used by more than just Siri - apps, shortcuts, etc can call/run those models for local inference.