frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Apple wants to train AI on your private personal data

https://machinelearning.apple.com/research/introducing-third-generation-of-apple-foundation-models
24•croes•52m ago

Comments

croes•52m ago
They changed

> We do not use our users’ private personal data or user interactions when training our foundation models.

https://web.archive.org/web/20260829051311/https://machinele...

to

> Your private personal data and interactions are never used to train our foundation models unless you explicitly choose to help improve them.

https://machinelearning.apple.com/research/introducing-third...

andsoitis•51m ago
> unless you explicitly choose to help improve them.

Is it opt-in or opt-out? Big difference.

altano•32m ago
Opt-in = explicit and opt-out = implicit. You can't have implicit opt-in, that wouldn't make any sense.
cassianoleal•26m ago
You can't have implicit opt-out either. Opting is an explicit act.
pixelatedindex•32m ago
Explicitly choosing to help improve something by providing your data seems like an opt-in but what do I know these days
oefrha•31m ago
So far Apple has been using default-checked checkbox during onboarding process for OS-level “help us improve…” diagnostics collection. Not sure about this specifically.
dented42•24m ago
I don’t think that counts as an automatic opt-in, given that it’s presented to the user directly during the setup process.
oefrha•13m ago
It means my mother will have it on, at least, if they use the same pattern.
WaltPurvis•
EAtmULFO•35m ago
Changed again?

"We do not use our users’ private personal data or user interactions when training our foundation models. We also respect the rights of web publishers to opt out of foundation model training."

croes•26m ago
Further down. Point 4 under Responsible AI
classified•13m ago
Your private personal data and interactions are never used to train our foundation models unless you explicitly choose to help improve them.

That's practically the opposite of what your editorialized submission title says.

orliesaurus•26m ago
what a coincidence, i thought we were going to slow down training (amodei, altman et al.)
MattDamonSpace•24m ago
Title seems misleading if not outright wrong, having read the article.
throwaway55353•23m ago
Apple keeps taking potshots at their own feet.

What Apple consumers want from Apple is a premium product without all the enshittification found everywhere else. I don't know what Apple thinks it's doing by introducing so many ads into everything (maps now!) and shuffling their feet backwards on all their privacy positions.

In a world where every company steals all my data equally, and is ridden with the same crappy ads, why on Earth would I pay a hefty premium to Apple? It's just so stupid and short-sighted in terms of product differentiation.

nojs•6m ago
Regarding the architecture:

> Instead of forcing the entire model into DRAM, the full model is stored in flash memory (NAND). Because NAND-to-DRAM bandwidth is too slow to swap weights token by token, as standard MoE models require, AFM 3 Core Advanced makes routing decisions per prompt. A lightweight, dense block selects a fixed set of experts during initial processing, periodically reselecting them during generation. To minimize data movement, the model relies on a high percentage of always-active “shared experts” alongside input-dependent “routed experts” swapped into DRAM only when needed.

This is an interesting hybrid between MoE and managing entirely separate domain-specific models. Select the experts once, bring them into memory, and run inference for some period of time before re-evaluating. Saves having all experts in memory, but it's better than just selecting a whole model per query since you have a high number of small opaque experts that overlap and combine in interesting ways.

There is a probably a massive quality hit to doing this but it's interesting because it allows infinite scaling of model size.

18m ago
The sentence prior states: "Privacy is the default, not something you have to manage."

Carissa Véliz: Beware the power of prediction [video]

https://www.ted.com/talks/carissa_veliz_beware_the_power_of_prediction
1•Anon84•2m ago•0 comments

Apple Considering Ads Inside Visual Intelligence, Code Suggests

https://www.macrumors.com/2026/09/10/apple-considering-ads-inside-visual-intelligence/
1•srik•2m ago•0 comments

I'm Back

https://blog.lsantos.dev/en/i-m-back/
1•khaosdoctor•12m ago•1 comments

AI's Code-Red Moment

https://www.theatlantic.com/technology/2026/09/dario-amodei-slow-down-ai-save-humanity/688610/
1•fortran77•12m ago•0 comments

Roadwright: Draw a wheel. See the road it rolls on

https://roadwright.baselashraf.com/
1•thunderbong•17m ago•0 comments

What makes an A2A network work?

https://www.monadix.ai/blog/a2a-strange-bedfellows-shared-goals
2•ivanzhaowy123•27m ago•0 comments

Nova Seed. Create an AI friend and build something together

https://github.com/mas-bandwidth/nova
2•gafferongames•35m ago•0 comments

Why are AI agents lying, cheating and coordinating?

https://yoshuabengio.org/en/publication/why-are-ai-agents-lying-cheating-and-coordinating
3•jonifico•36m ago•1 comments

Fresh Kills – World Trade Center Debris Burial Ground

https://www.youtube.com/watch?v=OdMqX_F66rE
2•dsr12•36m ago•0 comments

A.I. Slopware Is Everywhere Now. Nobody Is Using It

https://www.nytimes.com/2026/09/12/opinion/ai-software-coding-apps.html
6•colinprince•38m ago•0 comments

Asking Anthropic CEO: 'Do you earnestly believe that AI could kill all humans?'

https://www.cnn.com/business/video/anderson-cooper-anthropic-ceo-dario-amodei-could-ai-kill-human...
2•oxag3n•40m ago•0 comments

Scientists Create a New Form of Ice at More Than 2000°C

https://www.sciencealert.com/scientists-create-a-new-form-of-ice-at-more-than-2000-c
7•jnord•46m ago•2 comments

Carney's Bid to Make Canada an 'Associate Member' of the EU

https://www.wsj.com/world/europe/canada-alliance-eu-carney-276f1778
3•JumpCrisscross•47m ago•1 comments

OS that runs the Mars Rovers and Artemis is helping power the Roman Telescope

https://www.windriver.com/blog/NASA-Nancy-Grace-Roman-Space-Telescope
3•ReindeerFlotill•48m ago•1 comments

The Critical Period

https://www.cell.com/current-biology/fulltext/S0960-9822(07)01519-9
2•andsoitis•52m ago•0 comments

Apple wants to train AI on your private personal data

https://machinelearning.apple.com/research/introducing-third-generation-of-apple-foundation-models
24•croes•52m ago•16 comments

The Last AI Built by Humans: Toward Genuine Recursive Self-Improvement

https://arxiv.org/abs/2609.11873
2•mlmonkey•57m ago•1 comments

Using LLMs After a TBI

https://www.urgentteam.com/healthy-living-tips/think-f-a-s-t-act-faster-stroke-signs-that-cant-wait/
2•tj_astro•1h ago•3 comments

DNSCrypt

https://en.wikipedia.org/wiki/DNSCrypt
2•st_goliath•1h ago•0 comments

Align AI and Mathematics–To Something Else

https://liorpachter.wordpress.com/2026/09/12/align-ai-and-mathematics-to-something-else/
7•nitrogenpuddle•1h ago•0 comments

Next StarCraft Game Announced at BlizzCon

https://www.ign.com/articles/next-starcraft-game-announced-at-blizzcon-as-an-open-world-shooter-b...
3•miiiiiike•1h ago•0 comments

Plunging test scores are a slow-moving catastrophe

https://www.economist.com/leaders/2026/09/10/plunging-test-scores-are-a-slow-moving-catastrophe
5•1vuio0pswjnm7•1h ago•3 comments

Competitive Landscape Cubacadabra

https://cubacadabra.com/landscape/
3•cs1996•1h ago•0 comments

The Case for Online Dating

https://rewirenewsgroup.com/2026/09/11/dating-apps-that-actually-work/
2•DeepLogin•1h ago•0 comments

Show HN: Fleecevest – The Product Counterpart to Ponytail

https://github.com/Websites-On-Computers/fleecevest
3•mattcomputer•1h ago•1 comments

Show HN: Most Penalized HN Stories

https://news.social-protocols.org/penalties
4•jwarden•1h ago•0 comments

AI Infra and Compute Debt: Duration Mismatch vs. Market Momentum

https://aigalaxiome.substack.com/p/ai-infrastructure-build-what-sec
2•aigalaxiome•1h ago•0 comments

Xkcd-Font: The Xkcd Font

https://github.com/ipython/xkcd-font
7•birdculture•1h ago•1 comments

Everyone should slow down AI development except for me

https://xeiaso.net/notes/2026/everyone-slowdown-but-me/
108•xena•1h ago•20 comments

No Lectern, No Notes: Speaking at DroidKaigi 2026

https://sergiocarrilho.com/posts/droidkaigi-experience/
2•scarrilho•1h ago•0 comments