Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

I extracted the safety filters from Apple Intelligence models

github.com

Post reply on HN
Print view View original

361–370 of 455 posts

Re: I extracted the safety filters from Apple Intelligence models

#361

Post by chmod775 »

weinzierl wrote: post #336
chmod775 wrote: post #301

Earlier quoted context omitted.

> The OK gesture has always been very inappropriate in most parts of the world. No, it isn't, and especially hasn't been historically. The negative connotations are overwhelmingly modern. The areas where it is very inappropriate right now tally up to maybe 1 billion people*. That's pretty far from "most". For everyone else it is mostly positive, neutral, or meaningless. *Brazil, Turkey, Iran, Iraq, Saudi Arabia, Gree…

"No, it isn't, and especially hasn't been historically. The negative connotations are overwhelmingly modern." Maybe that is what Richard Nixon thought as well when he caused a little scandal using it in South America in 1950. In 1992 when the Chicago Tribune published " HANDS OFF" mentioning said episode the negative connotations still seemed to be in place[1]. In 1996 The New York Times stated "What's A-O.K. in the…

Care to add any country to the list then? Did I miss anything? Let's see if we can push it past half of the world's population, but I don't think we will.

> I can assure you in most parts using your thumb and index finger for a visual OK is not OK.

You're moving goal posts. Of course it doesn't just mean "OK" in some places.

What you actually claimed was "The OK gesture has always been very inappropriate in most parts of the world."

Which is plain wrong. In India for instance it can refer to "money", while in China it can nowadays also be seen as a distress signal when performed a certain way (thanks to Chinese social media popularizing that use). There's some ways you can mess this up, like making it seem you're attempting to bribe someone, or signalling you're in distress when you aren't, but in neither country the gestures are inherently anywhere near "very inappropriate" and both will even understand it as "OK" if you perform it correctly and in the appropriate context.

That's already almost 3 billion people, but let's say 2.5 billion because there's regional variations in both countries and I'm sure you could find some northern Chinese village that will take offense.

I can easily push the number of people to whom it is not inappropriate past 4 billion by adding smaller populations (Indonesia, Japan, western Europe, USA, Taiwan, South Africa, Kenya, Nigeria, ...), so your claim that "[it] has always been very inappropriate in most parts of the world" cannot possibly be true.

Re: I extracted the safety filters from Apple Intelligence models

#362

Post by m3kw9 »

1f60c wrote: post #329

It's pretty easy to understand why Apple doesn't want its models to reproduce racial slurs, but what’s wrong with "Boris Johnson?" (See, e.g., here: https://github.com/BlueFalconHD/apple_generative_model_safet... )

But allow hitler?

Re: I extracted the safety filters from Apple Intelligence models

#363

Post by mopsi »

weinzierl wrote: post #336
chmod775 wrote: post #301

Earlier quoted context omitted.

> The OK gesture has always been very inappropriate in most parts of the world. No, it isn't, and especially hasn't been historically. The negative connotations are overwhelmingly modern. The areas where it is very inappropriate right now tally up to maybe 1 billion people*. That's pretty far from "most". For everyone else it is mostly positive, neutral, or meaningless. *Brazil, Turkey, Iran, Iraq, Saudi Arabia, Gree…

"No, it isn't, and especially hasn't been historically. The negative connotations are overwhelmingly modern." Maybe that is what Richard Nixon thought as well when he caused a little scandal using it in South America in 1950. In 1992 when the Chicago Tribune published " HANDS OFF" mentioning said episode the negative connotations still seemed to be in place[1]. In 1996 The New York Times stated "What's A-O.K. in the…

That might have been the case decades ago. For example, in the USSR, various finger gestures usually implied something related to a penis and were considered extremely offensive. But that hasn't been the case since at least the early 1990s, when VCRs became widely available, people saw Hollywood movies for the first time and got used to westernized meaning of thumbs-up and OK gestures. Nowadays, when backing a truck towards a trailer, a thumbs-up would be taken as "good job" and an OK gesture (often paired with a kiss) as "exceptionally well done".

Re: I extracted the safety filters from Apple Intelligence models

#365

Post by thinkingtoilet »

troupo wrote: post #215
energy123 wrote: post #210

Earlier quoted context omitted.

This is the rhetorical tactic of false equivalence. State censorship by an autocracy with the objective of population control is not the same thing as a private company inside a democracy censoring their product to avoid bad press and maintain goodwill for shareholders. If you want solid proof that it's not the same thing, see all the uncensored open weights models that you can freely download and use without fear of…

> is not the same thing as a private company inside a democracy censoring their product to avoid bad press and Yet this private company has more power and influence than most countries. And there are several such companies. We already live in sci fi corporate dystopia, we just haven't fully realised it yet.

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.

Re: I extracted the safety filters from Apple Intelligence models

#366

Post by nedt »

1f60c wrote: post #329

It's pretty easy to understand why Apple doesn't want its models to reproduce racial slurs, but what’s wrong with "Boris Johnson?" (See, e.g., here: https://github.com/BlueFalconHD/apple_generative_model_safet... )

I think it's in there so you can't let it generate an email reply about how awesome peppa pig is.

Re: I extracted the safety filters from Apple Intelligence models

#367

Post by DamnInteresting »

hopelite wrote: post #334
junon wrote: post #274

Earlier quoted context omitted.

Also feels like some of these would match totally innocuous usage. "I'm overloaded for work, I'd be happy if you took some of it off me." "The client seems to have passed on the proposed changes." Both of those would match the "death regexes". Seems we haven't learned from the "glbutt of wine" problem of content filtering even decades later - the learnings of which are that you simply cannot do content filtering base…

This is a bigger issue, especially with Apple, than people may realize. I use iOS “Slide to Type”, aka swipe typing, and have noticed over time that among several other glitchy bad UX issues, there a clear heavy hand on what can be typed that way. I cannot recall all the specific patterns I have encountered that are basically impossible to write, some very similar in that they have a serious but also innocuous or fig…

> 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.

Re: I extracted the safety filters from Apple Intelligence models

#368

Post by jowea »

notarobot123 wrote: post #298
selfmodruntime wrote: post #264

Earlier quoted context omitted.

It's also a shining example of American puritanism. Asian models or those in Europe are far less censored.

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

At first I thought of advertisers, but that is not relevant here, right?

But maybe it's not just legal liability but bad press too.

Re: I extracted the safety filters from Apple Intelligence models

#369

Post by Spivak »

immibis wrote: post #321
selfmodruntime wrote: post #264

Earlier quoted context omitted.

It's also a shining example of American puritanism. Asian models or those in Europe are far less censored.

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. ( Qwen Mistral is French, but I have no idea what stuff would be censored in France)

If you ask the web UI it will divert, if you download and ask the model directly it will talk all day about it.

Re: I extracted the safety filters from Apple Intelligence models

#370

Post by weinzierl »

chmod775 wrote: post #361
weinzierl wrote: post #336

Earlier quoted context omitted.

"No, it isn't, and especially hasn't been historically. The negative connotations are overwhelmingly modern." Maybe that is what Richard Nixon thought as well when he caused a little scandal using it in South America in 1950. In 1992 when the Chicago Tribune published " HANDS OFF" mentioning said episode the negative connotations still seemed to be in place[1]. In 1996 The New York Times stated "What's A-O.K. in the…

Care to add any country to the list then? Did I miss anything? Let's see if we can push it past half of the world's population, but I don't think we will. > I can assure you in most parts using your thumb and index finger for a visual OK is not OK. You're moving goal posts. Of course it doesn't just mean "OK" in some places. What you actually claimed was "The OK gesture has always been very inappropriate in most part…

> I can assure you in most parts using your thumb and index finger for a visual OK is not OK.

>>You're moving goal posts. Of course it doesn't mean "OK" in many

I said the gesture is "not OK" to use (meaning inappropriate), not that it doesn’t mean "OK". Those are two different things. The gesture can mean OK in some places while still being not OK (inappropriate) to use in many others.

Also, I always said "parts of the world". You introduced population into the argument.

Post reply on HN