Apple Unveils Apple Intelligence

[OFaceSIG]

I think ultimately they’d end up doing that if AI sticks around. Plenty of people think it’s a fad and won’t grow much beyond its current capacity unless major tech breakthroughs happen. Apple may be playing the backfield on this one to see what happens. They aren’t usually cutting edge, but they rarely make mis steps.when it comes to tech. There are some, but they don’t have the grave yard of abandoned products like Google does.

Okay I can see the take a step back and wait and see angle. I guess only time will tell.

 

[DukenukemX]

I think ultimately they’d end up doing that if AI sticks around. Plenty of people think it’s a fad and won’t grow much beyond its current capacity unless major tech breakthroughs happen. Apple may be playing the backfield on this one to see what happens. They aren’t usually cutting edge, but they rarely make mis steps.when it comes to tech. There are some, but they don’t have the grave yard of abandoned products like Google does.

Apple is doing it because the letters AI seems to boost stock growth. As of right now, other than for artwork and helping people code, I don't see AI having much of a use outside it. What I do see AI being used for is to collect even more information on it's users. AI right now makes a lot of mistakes, and that's because AI depends on the people of the internet for answers. AI doesn't actually think, like the name would suggest, but it's really good at searching for information and presenting it in such a way that doesn't sound like a robot. It seems that Microsoft and now Apple are using their users directly to collect more accurate information so AI doesn't tell people to eat one rock a day or use glue on their pizza. Which is probably why Linux has been suggested as a replacement since this AI junk is being forced down peoples throats. I can only imagine the reason Elon Musk isn't happy about Apple using OpenAI is because it would help better train it. Especially since he's the owner of Twitter/X where iPhones can better train off of. Companies are spending Billions to better train their AI because it recommends putting glue on pizza, and it's getting to the point where it's scaring people to Linux. It's a fad that will crash harder than the dotcom bubble.

 

[sfsuphysics]

The feature, referred to as "A.I."

Yeah who didn't see that one coming when you first started to read the article? Now the Google side just needs to name theirs as "Android Intelligence"

 

[LukeTbk]

I can only imagine the reason Elon Musk isn't happy about Apple using OpenAI is because it would help better train it.

They work on advanced (and direct competitor to), people and company like him have actual reason to care about privacy issues.

Companies are spending Billions to better train their AI because it recommends putting glue on pizza,

That a bit unfair, that was AI that was purely resuming the top google search result and was being quite agnostic (i.e. it was not a answer based on its training) and was doing its task perfectly fine, would you ask the different AI they were not saying to do that at all,

I really doubt Apple will use anything for training data they say they won't.

 

[Aurelius]

Apple is doing it because the letters AI seems to boost stock growth. As of right now, other than for artwork and helping people code, I don't see AI having much of a use outside it. What I do see AI being used for is to collect even more information on it's users. AI right now makes a lot of mistakes, and that's because AI depends on the people of the internet for answers. AI doesn't actually think, like the name would suggest, but it's really good at searching for information and presenting it in such a way that doesn't sound like a robot. It seems that Microsoft and now Apple are using their users directly to collect more accurate information so AI doesn't tell people to eat one rock a day or use glue on their pizza. Which is probably why Linux has been suggested as a replacement since this AI junk is being forced down peoples throats. I can only imagine the reason Elon Musk isn't happy about Apple using OpenAI is because it would help better train it. Especially since he's the owner of Twitter/X where iPhones can better train off of. Companies are spending Billions to better train their AI because it recommends putting glue on pizza, and it's getting to the point where it's scaring people to Linux. It's a fad that will crash harder than the dotcom bubble.

View attachment 659199

The only people this will realistically scare toward Linux are the ones already on the cusp of switching. Some folks are nervous about AI, but few are so rattled that they'd actually ditch the platform they're using instead of... say, not using the features.

Remember that old "this is the year of Linux on the desktop" joke? It holds up precisely because the things that would actually get people to switch still aren't there, like their familiar apps, hardware ecosystems, and dramatic UI improvements.

 

[DukenukemX]

Yeah who didn't see that one coming when you first started to read the article? Now the Google side just needs to name theirs as "Android Intelligence"

What does the 'A' stand for ? Red Vs Blue quotes never get old.

View: https://youtu.be/An4gF0Gbpyw?si=QkMzmCt3lEW6jo3-

They work on advanced (and direct competitor to),

Which is why he's mad.

people and company like him have actual reason to care about privacy issues.

Everyone has a reason to care, but he's the owner of Twitter which is every companies wet dream to use for training their AI.

That a bit unfair, that was AI that was purely resuming the top google search result and was being quite agnostic (i.e. it was not a answer based on its training) and was doing its task perfectly fine, would you ask the different AI they were not saying to do that at all,

Again, AI doesn't actually think. All AI does is look at matches to what is more likely to answer your question. For example if you were to ask ChatGPT what the 21st letter is, it will answer confidently but also incorrectly. Tell it it's wrong and it will try again. AI is just not dependable no matter where you look. Even the use of it making images can often break consistency as parts of clothing will often change.

I really doubt Apple will use anything for training data they say they won't.

Not sure what Apple said, but often they will say they won't but legally they just won't share it with other companies.

The only people this will realistically scare toward Linux are the ones already on the cusp of switching. Some folks are nervous about AI, but few are so rattled that they'd actually ditch the platform they're using instead of... say, not using the features.

I don't disagree, but if you say it enough times then it'll happen.

Remember that old "this is the year of Linux on the desktop" joke? It holds up precisely because the things that would actually get people to switch still aren't there, like their familiar apps, hardware ecosystems, and dramatic UI improvements.

Or just wait for Microsoft and Apple to screw up badly enough that people will switch.

 

[LukeTbk]

Everyone has a reason to care

No obviously not, most do not care and have no reason too (at least went it come to a company like Apple knowing stuff about what they do)

Again, AI doesn't actually think.

This a complete different subject about the AI resuming top google search result correctly or incorrectly. You are some tech informed person, you do not have to fall for nonsense click-bait.

The nuance between thinking and what is the most likely answer to a question can become quite thin and irrelevant in most case. When AI came up with a new better CRISPR protein (that did not exist in nature), should we care if it was thinking or not ? When it predict meteo better than a 200 millions computer can using just a couple of cents of electricity ? Self-driving car causing less yearly accident, would they think or not ? No, it is artificially intelligence, like flying it reproduce the output of intelligence without being intelligent. Plane do not fly like bird, but if they go over obstacles in the sky, good enough, calculator don't do math like humans if the results are good, good enough.

Not sure what Apple said, but often they will say they won't but legally they just won't share it with other companies.

Federighi said “many” of the new AI features will run entirely on a device, without data leaving the phone. That’s generally an important security feature, reducing the risk that personal information will fall into someone else’s hands.

For other features that require more computing power, he said, Apple is setting up a new cloud-based system called Private Cloud Compute, where data could be processed for AI-enabled features but “never stored or made accessible to Apple.” He said the system would also be open to scrutiny from outside experts.

 

[Airbrushkid]

As we type on this forum Artificial Intelligence has already replaced humans. Human's are losing their jobs to Artificial Intelligence. Their restaurants running with out humans. But they need humans to buy from them. As time go's on there will be less humans on this planet.

 

[Lakados]

As we type on this forum Artificial Intelligence has already replaced humans. Human's are losing their jobs to Artificial Intelligence. Their restaurants running with out humans. But they need humans to buy from them. As time go's on there will be less humans on this planet.

Yet those robots need people to service them and keep tabs on them, they augment the labor market and change the needed skills to operate.
And you make the idea of fewer people sound bad, there's too many of us and despite the universe's attempts to get rid of us, we keep not going away.

 

[Airbrushkid]

Well if AI doesn't get rid of use. We humans will do it.

 

[Lakados]

Well if AI doesn't get rid of use. We humans will do it.

AI will destroy us, not because it wants to but because it is unable to tell the difference between euphemism and metaphor.
Pornhub and all these mafia movies all say humans love to be rubbed out, so this must be 100% accurate information as two conflicting resources both agree.

 

[Mchart]

Yet those robots need people to service them and keep tabs on them, they augment the labor market and change the needed skills to operate.
And you make the idea of fewer people sound bad, there's too many of us and despite the universe's attempts to get rid of us, we keep not going away.

Physical robots do. Most service jobs that exist entirely within a digital construct will be completely removed, and they don’t need much human oversight. Most everyone doing stuff like DB admin, cloud engineering, programming, etc AI will largely take those jobs.

 

[Airbrushkid]

What are you doing on Pornhub? No woman. I never in all these year have been to that site. Never a need. I've been married my whole life and have grand kids. And AI will end the human race. Don't naive to it.

AI will destroy us, not because it wants to but because it is unable to tell the difference between euphemism and metaphor.
Pornhub and all these mafia movies all say humans love to be rubbed out, so this must be 100% accurate information as two conflicting resources both agree.

 

[DukenukemX]

No obviously not, most do not care and have no reason too (at least went it come to a company like Apple knowing stuff about what they do)

I'm sure most people would care if they use their computers for professional use. I also don't trust Apple anymore than Microsoft or Google. Apple has been caught multiple times lying about what they do with your data and taken to court.

This a complete different subject about the AI resuming top google search result correctly or incorrectly. You are some tech informed person, you do not have to fall for nonsense click-bait.

I'm pointing out the many times AI has made mistakes. It being called AI is really relevant, just like Tesla's Autopilot is causing accidents.

The nuance between thinking and what is the most likely answer to a question can become quite thin and irrelevant in most case. When AI came up with a new better CRISPR protein (that did not exist in nature), should we care if it was thinking or not ? When it predict meteo better than a 200 millions computer can using just a couple of cents of electricity ? Self-driving car causing less yearly accident, would they think or not ? No, it is artificially intelligence, like flying it reproduce the output of intelligence without being intelligent. Plane do not fly like bird, but if they go over obstacles in the sky, good enough, calculator don't do math like humans if the results are good, good enough.

The problem is that these things already existed before. Calling it AI is just giving more credit to a stock market push. AI has already existed before as well, but again to what extent should we care?/

Federighi said “many” of the new AI features will run entirely on a device, without data leaving the phone. That’s generally an important security feature, reducing the risk that personal information will fall into someone else’s hands.

For other features that require more computing power, he said, Apple is setting up a new cloud-based system called Private Cloud Compute, where data could be processed for AI-enabled features but “never stored or made accessible to Apple.” He said the system would also be open to scrutiny from outside experts.

As a reminder, Apple's new iOS update has brought back deleted photos. Do you really trust them with more sensitive data?

 

[LukeTbk]

I'm pointing out the many times AI has made mistakes. It being called AI is really relevant, just like Tesla's Autopilot is causing accidents.

But some of those examples were not, was the AI resuming top google research correctly or made mistake (in the viral example named, the AI was correct, the Top google result were bad, but I imagine google result use Ai a lot to rank but they also use commercial interest)

The problem is that these things already existed before. Calling it AI is just giving more credit to a stock market push. AI has already existed before as well, but again to what extent should we care?/

What think ? No a small laptop was not able in less than 2 minutes to predict the meteo in 14 days for any gps location only using RSS feed of the previous world meteo information better than a 200 millions supercomputer and no we were not finding new CRISPR protein just by giving a list of example of existing protein to a computer, self-driving were always called artificial intelligence, this is not some change.

The change is much more in the other direction, stuff stop to be called AI, like a computer that play chess went from being called to stop to be, NPC in half-life was called impressive AI and now seem to stop to be, voicemail that understand your voice was AI and seem to stop to be according to some, Ai always existed since computer existed and we tend to stop calling what they do AI once they get good at it.

Do you really trust them with more sensitive data?

Yes, I do my banking from my iPhone. It is much easier to never save data, to have extremely secure level of never your loose your file backup (with a recently deleted category and so on) like an Apple Cloud and not delete them correctly, which sound like a technical error (seem to have been from the phones having putt picture in their files app and the user having only deleting them from their pictures app).

 

[DukenukemX]

But some of those examples were not, was the AI resuming top google research correctly or made mistake (in the viral example named, the AI was correct, the Top google result were bad, but I imagine google result use Ai a lot to rank but they also use commercial interest)

That's how 'AI' works, by looking for something that already exists. The problem is that AI doesn't know that using glue to hold cheese onto a pizza is a bad idea. This is the problem with AI in that it depends on existing information to be able to regurgitate it. That existing information just so happens to be from a reddit post 11 years ago from a 12 year old. How about when ChatGPT that was trained on data to be a lawyer and was citing cases that were made up? It's not infrequent that ChatGPT will make something up if it doesn't know the proper answer.

What think ? No a small laptop was not able in less than 2 minutes to predict the meteo in 14 days for any gps location only using RSS feed of the previous world meteo information better than a 200 millions supercomputer and no we were not finding new CRISPR protein just by giving a list of example of existing protein to a computer, self-driving were always called artificial intelligence, this is not some change. Google made their own AI chips long before this ChatGPT fueled AI craze.

Does a small laptop do that now? Do we not count laptops with RTX 2060's or 3060's? We were already predicting proteins long ago. We did stuff with Folding at Home that also found new proteins. Google's been using AI for self driving as far back as 2009. It's not like Google's Waymo is doing that much better with all the car crashes lately.

View: https://youtu.be/K511Jnl4x2Q?si=cI1mqR_3U8Rr6jBY

The change is much more in the other direction, stuff stop to be called AI, like a computer that play chess went from being called to stop to be, NPC in half-life was called impressive AI and now seem to stop to be, voicemail that understand your voice was AI and seem to stop to be according to some, Ai always existed since computer existed and we tend to stop calling what they do AI once they get good at it.

We stopped calling them AI when people's eyes started to roll up. We begin to look at it for what it really is.

Yes, I do my banking from my iPhone.

Some reason people like to let Apple manage their money. You are being properly rewarded for that poor decision. You underestimate the amount of times Apple has screwed up.
https://dailyhodl.com/2023/06/03/af...stomers-are-struggling-to-access-their-money/
"After Depositing $1,000,000,000, Apple Banking Customers Are Struggling to Access Their Money"

 

[Airbrushkid]

I would never use a cell phone to do banking! Not to bright doing it. Unless your rich enough to lose money.

 

[Mchart]

I would never use a cell phone to do banking! Not to bright doing it. Unless your rich enough to lose money.

Yes, I’m sure your windows 7 machine on your Asus router home network is more secure.

 

[Airbrushkid]

WOW you speak before you know. I have a 6 minute walk to my bank! I only do my banking at my bank in person!!!

Yes, I’m sure your windows 7 machine on your Asus router home network is more secure.

 

[DukenukemX]

Elon Musk says that Apple will be handing your data over to OpenAI. Don't know where people get the idea that Apple won't.
https://x.com/elonmusk/status/1800269249912381773

 

[Airbrushkid]

View: https://www.youtube.com/watch?v=pMX2cQdPubk

 

[Lakados]

I'm sure most people would care if they use their computers for professional use. I also don't trust Apple anymore than Microsoft or Google. Apple has been caught multiple times lying about what they do with your data and taken to court.

I'm pointing out the many times AI has made mistakes. It being called AI is really relevant, just like Tesla's Autopilot is causing accidents.

The problem is that these things already existed before. Calling it AI is just giving more credit to a stock market push. AI has already existed before as well, but again to what extent should we care?/

As a reminder, Apple's new iOS update has brought back deleted photos. Do you really trust them with more sensitive data?

iOS doesn’t delete photos, when you delete a photo it gets highly compressed and moved to a storage container for deleted photos. To completely delete a photo you need to delete them from the deleted photos repository just like emptying the trash can.

Apple got into some legal heat back in the day because somebody accidentally deleted important photos and they argued Apple didn’t have enough popups in place to prevent it from happening. They settled and added the feature.
It’s what happens when old windows users move to a new product, they expect the same degree of hand holding and 30 prompts to confirm a button click.

But if an update needs to make changes to the repository it’s safer to decompress them add them back and make the changes than it is to risk destroying them and getting “iOS destroyed my photos” blasted in the headlines and a class action.

That said it should be relatively simple to flag them in the process and add a routine to them scan for the flag and put them back to the deleted pictures bucket.

 

[cpufrost]

83 cycles on my 1TB 15PM here. Got it on release day and it stays charged at 100%.
I've had other iPhones that were down to 93% in 6 months. Not sure about the battery capacity rating. Some worry over this like the end of the world is coming.
This whole AI thing seems like a gimmick anyhow. I mean what is the device/system actually learning? I can train a puppy to stop pissing on the floor in a few weeks. But my phone can't even figure out how to properly auto correct texts? Just the other other day it changed Sarah to Satan FFS!

 

[Airbrushkid]

So you got it from a store?

 

[LukeTbk]

That's how 'AI' works, by looking for something that already exists.

that a misunderstanding, the meteo of the futur does not already exist, the new CRISPS protein it came with didn't exist in nature or human literature, not sure if extrapolating or interpolating has difference for it or good word

The problem is that AI doesn't know that using glue to hold cheese onto a pizza is a bad idea.

Depends what you mean by knowing, chatgpt told me it was a bad idea (so was eating rocks) ,but I am not sure "know" is the good word. The AI jobs of resuming google top results was not trying to know anything, was trying to resume google search top result and did it

How about when ChatGPT that was trained on data to be a lawyer and was citing cases that were made up? It's not infrequent that ChatGPT will make something up if it doesn't know the proper answer.

The focus on the click-bait non sense.... yes those LLM often make stuff up (which contradict quite directly the only repeating-copy stuff that already exist part, that many people still hold in their head at the same time)

Does a small laptop do that now? Do we not count laptops with RTX 2060's or 3060's?

What do you mean, the small laptop example has a GPU, yes small laptop. Why would we not count laptop with RTX 2060 ? What different about them.

We did stuff with Folding at Home that also found new proteins. Google's been using AI for self driving as far back as 2009.

Yes AI ? AI existed since computer existed, yes we used giant compute for very little result versus what Alphafold his doing now in comparison. Humans we doing 2-3-4 years work to find one protein, alphafold do hundreds of thousands in weeks: The AlphaFold Database has expanded its collection of predicted protein structures by over 714 times.
https://deepmind.google/discover/bl...and-more-accurate-global-weather-forecasting/

Better result from 2 minutes on a simple laptop versus hours on a supercomputer.

It is not like neural network and machine learning has not been used for a very long time in sound-imagery and so on, but the transformer revolution just made a step in the function to how big and good they could get.

"After Depositing $1,000,000,000, Apple Banking Customers Are Struggling to Access Their Money"

I do not have any apple wallet or anything like that, but I have my banks app on my phone and use them, because I have a giant amount of confidence on the device and Apple.

We begin to look at it for what it really is.

That really not if, if, there

Alpha go type of Ai learned to play go or chess without any human intervention or even watching human games, it is all synthetic training data, I am not sure you understand the subject here, do you really think that Alphazero playing chess is just a bunch of if coded by someone ?

Do you think that a system with a bunch of IF, make it so that it can not have artificial intelligence ?

 

[SomeGiantIdiot]

Heh. Apple (3.33T) over took Microsoft (3.27T) and is now the most valuable company in the US.

 

[RanceJustice]

The partnership between Apple and OpenAI is unfortunate from an exposure, branding, legitimacy angle as it s two companies taking AI/LLM the wrong direction. As an aside, every time I say or write that company's name I want to gouge my eyes out, given its basically some sort of Orwellian newspeak. "OpenAI" is nearly entirely focused on proprietary LLMs, proprietary training data, and SAAS levels of access to be sold to everyone else; there is nothing "Open" about any of this and they have a massive vested profiteering interest in locking it down, pulling the ladder up to stop competitors etc. So now we have two companies that are known for proprietary "the only way is Our Way (tm) " approaches , bolstered by "REVOLUTIONARY MAGICAL" wiz-bang marketing to play well to people who may know or care very little about the underlying technology or how they're being carried in a direction that wil be contrary to everything but these companies' bottom lines.

If we want a future where the benefits of AI / LLM are widely distributed throughout society, we need to act now. We need to demand that any serious usage (especially government funded and/or used for public works in any way, as well as private businesses when they are contracted to work in these spaces or just end up doing something impacting the public etc) requires self-host capable open source models, open training data in open formats, and more. I haven't even gotten into things like updating our now 20+ years out of data privacy laws that are woefully insufficient for the 90s era of the Internet much less the arrival of advertising/data mining, social media and now AI. However, to stick to AI, if we don't move soon to be sure that models and data and their formats are open, we're moving towards the cyberpunk dystopia where a handful of megacorps will be the only ones with top tier AI and everyone and everything else will have to contract for access to their technology, software as a service style. For the vast majority of people this will mean access to a sanitized, limited model that cannot do anything that would be seen as less marketable, and even if it started giving people bad advice the companies will be protected by the "its a trade secret" blackbox to look into their training data or other functions. Sure, there may be a few "hobbyist" models but they'll be vastly underpowered and less sophisticated compared to the billion dollar companies "serious" models, not to mention attempts to "regulate reponsible usage", mostly upheld by these megacorps just so happen to put onerous restrictions financial and otherwise on any open model and/or competitor who' not a venture capital funded megacorp; the aformentioned "pulling the ladder up", and being championed by misguided and self interested types with hate for AI (the whole social media "artists against AI" nonsense)

We don't have a lot of time to get this right and be sure that the benefits of LLM/AI aren't only available to elite tech companies and other megacorps who can make use of it to further ensconse their wealth and power and put everyone else in a more subjugated stance, begging for scraps of the technological future while wondering why things are getting worse. There are so many things I've not even touched on yet, but Apple and the so-called OpenAI partnering with their current operating dynamic doen't bode well for anything but their shareholders, damn the rest.

 

[LukeTbk]

As an aside, every time I say or write that company's name I want to gouge my eyes out, given its basically some sort of Orwellian newspeak. "OpenAI" is nearly entirely focused on proprietary LLMs, proprietary training data, and SAAS levels of access to be sold to everyone else; there is nothing "Open" about any of this and they have a massive vested profiteering interest in locking it down, pulling the ladder up to stop competitors etc. So now we have two companies that are known for proprietary "the only way is Our Way (tm) " approaches , bolstered by "REVOLUTIONARY MAGICAL" wiz-bang marketing to play well to people who may know or care very little about the underlying technology or how they're being carried in a direction that wil be contrary to everything but these companies' bottom lines.

It is for sure a bit that, but the claimed open goals and achievement are not fully false either.

ChatGPT 4-o, the text version, is completely free to use on the website, arguably the best and fastest outthere, free to use, to everyone, that quite extraordinary and considering the cost to make and run them that would be hard to do without charging money for the commercial uses of them.

And without an public and free to use interfacing to GPT-3 that they made, how much of the world would be taking attention to virtually the exact same tech being made at Google ? Behind doors and slowly tricking without ever making a "bang"

 

[RanceJustice]

It is for sure a bit that, but the claimed open goals and achievement are not fully false either.

ChatGPT 4-o, the text version, is completely free to use on the website, arguably the best and fastest outthere, free to use, to everyone, that quite extraordinary and considering the cost to make and run them that would be hard to do without charging money for the commercial uses of them.

And without an public and free to use interfacing to GPT-3 that they made, how much of the world would be taking attention to virtually the exact same tech being made at Google ? Behind doors and slowly tricking without ever making a "bang"

"Free to use" isn't really the point here; its "free" in the same way online games with the most predatory monetization are often "free to play" or Facebook is "free" to use as well; none of these things are "open" or for the benefit of anyone but the for profit entity that creates them. Its more of an advertisement than anything especially in the early days , not to mention a condition of that use is to basically contribute your usage and data for them to use as they wish. We've also see them begin to restrict the model's capabilities for what's available for free, sanitize inputs and outputs unless you know a rather sophisticated jailbreak whack-a-mole prompt engineering, and prohibiting access to the latest types except to those even individuals who are paying. For instance, ChatGPT Plus or whatever its called subscription (the price has increased at least once significantly as I recall), and DALL-E is not available at all directly unless you pay and/or go through someone like Microsoft who made their CoPilot/Bing Create or whatever they call it and thus provide data to both OpenAI and MS etc.

Ultimately though, the issue is that even using it for free is at the pleasure and benefit of the company and its all happening with their proprietary everything, top to bottom. What we need is something like Stable Diffusion, where the model and training data are truly open source/spec, can be self-hosted, iterated and added to in open ways, and built in a distributed fashion. There could still be plenty of ways to fund (both public and private) such continued development with an open approach to AI, and profit to be made from its use , but it would also limit the amount of difference between the haves and have nots in the AI space, to keep exploitationto a minimum. Anything else is tantamount to handing yet another force multiplier to billionaire megacorps to enrich themselves at the cost of everyone and everything else.

 

[LukeTbk]

We've also see them begin to restrict the model's capabilities for what's available for free

And we have seen the complete opposite as well, the free version capacity and speed exploded in the last month and much faster than any paid version was in many ways.

Ultimately though, the issue is that even using it for free is at the pleasure and benefit of the company and its all happening with their proprietary everything, top to bottom.

If the rumours of over 1 trillions weight for their models are true, open or close, how many companies in the world could run it ? How could they have spent that amount of money in a free and open model, everything is a compromise, always there never solutions. Impossible to know, but it would be a good bet that there existance has made things way more open than what the Google and others version would be now.

I am sure I remember a domain, where the free part exploded that much, small 3-7Billions models are getting really good, there even talk of upcoming 1 or 2bits, replacing the 4 bits, what will be needed to train and run model of today capacity could be quite accessible, quite fast.

 

[DukenukemX]

That really not if, if, there

Alpha go type of Ai learned to play go or chess without any human intervention or even watching human games, it is all synthetic training data, I am not sure you understand the subject here, do you really think that Alphazero playing chess is just a bunch of if coded by someone ?

Do you think that a system with a bunch of IF, make it so that it can not have artificial intelligence ?

I don't know, and it's not something that anyone has explained yet too me. So far the TPU's are using int8, which feels like going backwards. There's even talk using int4. Technically everything a computer ever does is a bunch of if else statements, but AI is likely using tables of data and shifting through them until it finds a correct function that matches the data the nearest, and repeats over and over. Which still sounds like if statements with less steps. All software is a bunch of conditional statements, but I just don't know how 'AI' is handled. Whatever the case is, it's still just software that sometimes works really well and sometimes really doesn't. Imagine if turning on your PC might boot into a MacOS or might boot into TempleOS. You don't want that level of reliability with your software.

 

[LukeTbk]

I don't know, and it's not something that anyone has explained yet too me. So far the TPU's are using int8, which feels like going backwards. There's even talk using int4.

int4 has been out for a while (lot of model performance gain on Blackwell come from them handling int4 better), int2 is what they're even talk now, even just 3 distinct possible weight at Microsoft:
https://dataconomy.com/2024/03/06/microsofts-1-bit-llm-advantages/
With a 1-bit LLM, each weight can only take on one of three values: -1, 0, or 1.
https://arxiv.org/pdf/2402.17764v1

The better performance for less cost you get, you could run 70B models on a 12GB 4070

You don't want that level of reliability with your software.

Depend on what your software do and in exchange for what, if I am lays and want to identify is the potato chips look good enough (color, shape) to go in the bag, I could be ok with an AI removing 2% too many potatoes specially if the coded algorithm is worst at it, anything video game of course it is not an issue if a mistake occurs, finding the best path often does not even a real best objective answer you could come with, but you do not want to let an LLM do math without it using an actual calculator.

Imagine upscaling an image, understanding voice to become text, the notion of being 100% reliable by following an algorithm cannot exist to start with, is this a stop sign or not in this camera feed, would I have a good chance of finding petrol according to those 3 terabytes of seismic, sound, etc.. data of the region. Computing is now filled of answer of the type, hum I would say 85% chance than yes regardless of how you would do it, in a way a basic Excel spreed sheet is not.

Technically everything a computer ever does is a bunch of if else statements,

On GPU you try to avoid if/else (it is one of the first thing you discover when you go from CPU coding to GPU coding they are so parallelizing everything that they will often do both, the things in the if true and in the else at the same time, they unloop loops aggressively), I am not sure if a computer doing the raster of a 3d scene or playing an mpeg2 is best imagined by a list of if/else in series, versus multiplying matrix in parallel.

 

[Teenyman45]

View: https://youtu.be/VoK0KdUlwoY

 

[DukenukemX]

Depend on what your software do and in exchange for what, if I am lays and want to identify is the potato chips look good enough (color, shape) to go in the bag, I could be ok with an AI removing 2% too many potatoes specially if the coded algorithm is worst at it, anything video game of course it is not an issue if a mistake occurs, finding the best path often does not even a real best objective answer you could come with, but you do not want to let an LLM do math without it using an actual calculator.

Imagine upscaling an image, understanding voice to become text, the notion of being 100% reliable by following an algorithm cannot exist to start with, is this a stop sign or not in this camera feed, would I have a good chance of finding petrol according to those 3 terabytes of seismic, sound, etc.. data of the region. Computing is now filled of answer of the type, hum I would say 85% chance than yes regardless of how you would do it, in a way a basic Excel spreed sheet is not.

This is more media related than anything else. I certainly don't have a problem with machine learning or AI being used to enhance art or audio, because in the end it's in the hands of a human. If AI gives a person 6 fingers then the artist could simply correct the mistake. We've already seen some pretty amazing things come from this, but how does this translate into what end users can experience? If you're a professional then you're not going to be using a TPU for generating art, but most likely an Nvidia GPU. Are end users going to get the software needed to make use of these new upscaling techniques for free? Can the TPU keep up with the GPU for gaming?

On GPU you try to avoid if/else (it is one of the first thing you discover when you go from CPU coding to GPU coding they are so parallelizing everything that they will often do both, the things in the if true and in the else at the same time, they unloop loops aggressively), I am not sure if a computer doing the raster of a 3d scene or playing an mpeg2 is best imagined by a list of if/else in series, versus multiplying matrix in parallel.

The If statement thing is a meme on the internet meant to demystify the voodoo around AI. Necessity is the mother of invention but so far AI is an invention that is looking for a necessity. Look at the crap Microsoft has created for Windows 11 that nobody wants, which in itself has created memes about switching to Linux. Meanwhile Apple is chasing the same pipe dream to better enhance Siri, while Google's Assistant has been better since forever. In most cases, Google Assistant was equal to ChatGPT. Google Assistant was already running machine learning, but when Google tried to "enhance it", we had people ready to eat rocks. One rock per day, as recommended by most doctors. Google were so preoccupied with whether they could, they didn't stop to think if they should. Now we have Apple ready to shoehorn AI into everything they make? I guess it can't be as bad as Recall... right?


View: https://youtu.be/Gyhn7D3y8X4?si=78adCpvYr7IfkbpM

 

[LukeTbk]

If you're a professional then you're not going to be using a TPU for generating art, but most likely an Nvidia GPU.

depend on if you are using google cloud it could be a TPU, I am really unsudre how it will shake up (or how it would matter even a little bit here, the nuance between those 2 scenario seem close to nill), what does it change if it is on a google TPU or an nvidia GPU ?

Necessity is the mother of invention but so far AI is an invention that is looking for a necessity.

This a bit of a strange statement, how would voice to text work, how would people work on self driving car, Ai was in large part driven by the gaz industry problem of looking for oil cost giant amount while the giant amount of seismic data use to try result before spending the money is hard for human to handle and understand.

Transformer made AI so powerful and there so much capitals flowing that yes, many people are trying to find stuff you could do with it, but from youtube algorithm suggestion, image recognition, to mRNA vaccine, face recognition, AI has been used in most field for a while now.

. Google Assistant was already running machine learning, but when Google tried to "enhance it", we had people ready to eat rocks.

You are turning around here, do you not understand than an AI was resuming google top search result, that contained that joke, it was not an google assistant saying to someone to do that.

Google were so preoccupied with whether they could, they didn't stop to think if they should.

This is a complete reverse of google strategy, like you say google was (and could still be by far) the most advanced company in the world in LLM and AI in general with probably the best data, why did they launch this so long after Bing, because they thought long for years if they should, google search was possible the best money printing machine in the history of humanity, should they change that model is something they thought long and hard and would have never do if not push.

The If statement thing is a meme on the internet meant to demystify the voodoo around AI.

Seem to be from people quite not understanding how it works, there reason a model does not give the same answer to the exact same question

 

[Mazzspeed]

Remember that old "this is the year of Linux on the desktop" joke? It holds up precisely because the things that would actually get people to switch still aren't there, like their familiar apps, hardware ecosystems, and dramatic UI improvements.

Yeah, this is a fairly exaggerated point of contention regarding the average user. I'm a somewhat advanced user, and everything I need is available cross platform.

When it comes to MS Office - I simply don't need it, I don't desire it, I'm certainly not paying an ever increasing subscription for it or the excessive cost in relation to buying it outright. Even if I ran Windows, I wouldn't use MS Office for the exact same reasons mentioned. When it comes to business, I've personally set up a number of businesses that use Google's suite just fine - In fact when it comes to collaboration, they love it.

I haven't thought the Linux desktop was a joke for a number of years now, furthermore it just keeps getting better. When people state that the Linux desktop isn't there yet, I look at my own desktop, my installed software, and wonder just what it is I'm supposed to be missing. I've stated it before, I'll state it again - I never find myself sitting in front of the PC wishing I was running Windows, in fact the reality is the polar opposite.

But I digress, back to the topic of AI that's anything but intelligent.

 

[Mchart]

Yeah, this is a fairly exaggerated point of contention regarding the average user. I'm a somewhat advanced user, and everything I need is available cross platform.

When it comes to MS Office - I simply don't need it, I don't desire it, I'm certainly not paying an ever increasing subscription for it or the excessive cost in relation to buying it outright. Even if I ran Windows, I wouldn't use MS Office for the exact same reasons mentioned. When it comes to business, I've personally set up a number of businesses that use Google's suite just fine - In fact when it comes to collaboration, they love it.

I haven't thought the Linux desktop was a joke for a number of years now, furthermore it just keeps getting better. When people state that the Linux desktop isn't there yet, I look at my own desktop, my installed software, and wonder just what it is I'm supposed to be missing. I've stated it before, I'll state it again - I never find myself sitting in front of the PC wishing I was running Windows, in fact the reality is the polar opposite.

But I digress, back to the topic of AI that's anything but intelligent.

Linux isn’t particularly good if you’ve got an enterprise running certain desktop apps. Namely office. It’s passable in an environment where most of what workers need are all web apps where the OS doesn’t matter.

I’m mostly Mac at this point. I get an OS that largely functions under the hood close to Linux, I get support for things like Office, and I’m nearly there completely ditching my windows gaming rig now that more AAA games are increasingly Mac native.

 

[DukenukemX]

depend on if you are using google cloud it could be a TPU, I am really unsudre how it will shake up (or how it would matter even a little bit here, the nuance between those 2 scenario seem close to nill),

If you're generating artwork then I doubt you'll be using Google's TPU. I think Google's TPU's are specialized for search results anyway. Also, Google's first TPU was built in 2015, which already predates all this.

what does it change if it is on a google TPU or an nvidia GPU ?

Accessibility and power I'd imagine.

This a bit of a strange statement, how would voice to text work, how would people work on self driving car,

We've had voice to text since the Windows XP era, at least. As for self driving, we've had companies that sometimes depend on people for it. Remember Amazon's Go store where people checked out just by picking up things? "Though it seemed completely automated, Just Walk Out relied on more than 1,000 people in India watching and labeling videos to ensure accurate checkouts."

View: https://youtu.be/FVKJ-GKSPX4?si=BczWDakLB2LJH9wF

Transformer made AI so powerful and there so much capitals flowing that yes, many people are trying to find stuff you could do with it, but from youtube algorithm suggestion, image recognition, to mRNA vaccine, face recognition, AI has been used in most field for a while now.

The YouTube algorithm is infamous and nobody likes it. Imagine recognition and face recognition is hated. mRNA vaccine I have no idea, but it's been around for a long time.

You are turning around here, do you not understand than an AI was resuming google top search result, that contained that joke, it was not an google assistant saying to someone to do that.

Yea but what classifies an algorithm as an AI? It's been here but yet now the world is crazy for AI? Amazon has had AI running their store and suggesting products to people for more than 10 years.

This is a complete reverse of google strategy, like you say google was (and could still be by far) the most advanced company in the world in LLM and AI in general with probably the best data, why did they launch this so long after Bing, because they thought long for years if they should, google search was possible the best money printing machine in the history of humanity, should they change that model is something they thought long and hard and would have never do if not push.

Sadly, Google search is getting unusable. Gets any worse and I may have to look for alternatives.

Seem to be from people quite not understanding how it works, there reason a model does not give the same answer to the exact same question

I'm one of those people. When people don't know how something works that can destroy your life if used maliciously, then people will have a tendency to avoid it.

 

[LukeTbk]

If you're generating artwork then I doubt you'll be using Google's TPU. I think Google's TPU's are specialized for search results anyway. Also, Google's first TPU was built in 2015, which already predates all this.

What all this mean ? The idea of computer neural network are old has computer (started during WW2), deep learning in the 60s, yes the 2017 Transformer from Attention is all you need revolutionary paper step boost in them is just one of the many evolution. Google has probably always been a couple of year ahead of everyone and could still be if not of OpenAI giving us an idea of where they are.

Yea but what classifies an algorithm as an AI?

There no rules and that change all the time, in the past human made algorithm were called AI (think Half-life NPC or the first chess playing computer) not much anymore.

Not it is mostly reserved to the non-algorithmic part black box system that were trained via some method of machine learning, we showed 10,000 different video (with different lighting, angle, size, quality) of a Coca-cola can to the system, 10,000 videos where there can looking but not really can object in them, telling them the answer, now if we show it an imagine it return a percentage of chance that it contains a coke can, humans did not tell it how and vaguely knows how it does, you cannot follow an algorithm going on, billions of weight neural network firing that will not necessarily give the exact same answer to the exact same image.

It's been here but yet now the world is crazy for AI?

The public at large, closer world has been crazy for AI since alphago at least (2015), look at Nvidia moves since, what the world think here without being 100% irrelevant, can of is, the only think it does it create giant investment and make progress really fast, but why do you care particularly about that ?

The transformer revolution of 2017, completely changed how big the neural network can get, the context window, like a step function and opened door to things a lot of people thought impossible, openAI effort to start a conversation around them exploded when they made GPT 3 public

I'm one of those people. When people don't know how something works that can destroy your life if used maliciously, then people will have a tendency to avoid it.

This is quite the dramatic view of DLSS or the system that give you direction (google map, etc...), once Ai does something well people will stop to call AI, soon if not already a computer that hear voice and make a text transcript of it people will stop calling that AI, in less than 10 years, computer hearing Italian and translating it outloud in Russian people will stop calling that AI, it will just be what computer do. The humans robots that actively training and taking decision, humanless car driving, war drone shooting without human intervention to still work under an EMP defense, will be the type of stuff they will still freak out about for a while. And after a while, young people will freak out that humans were killing 40,000 peoples on the road in the US every year instead of much safer robot car.

Remember how much we freaked out the first time google completed the word you started to type back in the days, people completely freakout when facebook came up with a new feeds of your own contacts contents now it is just normal, computer suggest a new video based on what you liked, now just normal.

 

[DukenukemX]

This is quite the dramatic view of DLSS or the system that give you direction (google map, etc...), once Ai does something well people will stop to call AI, soon if not already a computer that hear voice and make a text transcript of it people will stop calling that AI, in less than 10 years, computer hearing Italian and translating it outloud in Russian people will stop calling that AI, it will just be what computer do. The humans robots that actively training and taking decision, humanless car driving, war drone shooting without human intervention to still work under an EMP defense, will be the type of stuff they will still freak out about for a while. And after a while, young people will freak out that humans were killing 40,000 peoples on the road in the US every year instead of much safer robot car.

This is what I mean, in that what is an algorithm and what's an AI? AI is an algorithm, but not all algorithms are AI. What we're talking about here are things like OpenAI and ChatGPT which can being used to track data to either better train their machine learning algorithms or just have handy for reasons unknown to us. It's not just that 'AI' isn't working perfectly, but because we know that for this to work well it needs to feed off our data. DLSS isn't feeding back data to Nvidia, at least I hope it doesn't. Google maps does which was already an issue since Google was already profiting off people freely giving them data to better enhance their product without paying anyone for said data. If anyone knows what you did last summer, it would be Google.


View: https://youtu.be/nGIpdiQrFDU?si=IcpBtfgj1OpIK5rE

View: https://youtu.be/ek2yOqAIYuU?si=Ku9atGDg6-7wUo5I