Artificial Intelligence thread

[tphuang]

Your example above is mainly talking about B2B, whereas I'm talking about B2C. Imagine five years from now, or less, you go to your toy store in Spain, and parents buy their kids a little robotic, you know, teddy bear that makes API calls to multimodal ChatGPT 5.0 as an Omni. This is the type of world that could happen within a short order, now with Apple making this historical partnership with OpenAI, instantly putting OpenAI in basically half of the world's smartphones, and people all over the world are now unwittingly on America's AI platform, not even by choice.

why would they make API calls to GPT 5.0? The manufacturers of toys and robot stuff are all Chinese companies. They will use whatever the Chinese company hooks up to.

openAI isn't getting access to half of world's smartphones. iPhone's market share globally is maybe 15%. Keep in mind that Apple cann't use openAI in China, so the % is even lower than that.

On the other hand, people around the world that buys BYD cars, Xiaomi products and Honor/Vivo/Oppo phones are all having their data go into Chinese LLMs.

Similarly, anything you do with facebook or instagram goes to Meta. Anything you do with google searches go to Alphabet. Anything you do on douyin or TikTok goes to Bytedance.

if you want to make the argument that OpenAI will be getting additional user data that it previously did not access, okay that's fine. We will see how much data Apple provides to OpenAI. But there are a lot of additional data out there that only selected private companies have access to.

In fact, the top problem that OpenAI and others are running into is that they are running out of quality data.

And beyond that, you have not demonstrated at all what EUV has anything to do with this

 

[9dashline]

why would they make API calls to GPT 5.0? The manufacturers of toys and robot stuff are all Chinese companies. They will use whatever the Chinese company hooks up to.

openAI isn't getting access to half of world's smartphones. iPhone's market share globally is maybe 15%. Keep in mind that Apple cann't use openAI in China, so the % is even lower than that.

On the other hand, people around the world that buys BYD cars, Xiaomi products and Honor/Vivo/Oppo phones are all having their data go into Chinese LLMs.

Similarly, anything you do with facebook or instagram goes to Meta. Anything you do with google searches go to Alphabet. Anything you do on douyin or TikTok goes to Bytedance.

if you want to make the argument that OpenAI will be getting additional user data that it previously did not access, okay that's fine. We will see how much data Apple provides to OpenAI. But there are a lot of additional data out there that only selected private companies have access to.

In fact, the top problem that OpenAI and others are running into is that they are running out of quality data.

And beyond that, you have not demonstrated at all what EUV has anything to do with this

Intelligence (AI layers) will be the ultimate value add... China can make the physical products but the AI layer will be running US AI... just like Oppo, etc are Chinese branded phones but run US based Google andriod .... facebook rayban glasses are also made in China, but run meta AI as its brains

The Robosen Optimus Prime comes to mind, its an awesome transformer toy for $999 , made in China but by a US company... if they integrate it later with vision, audio, LLM, etc it will almost certainly be with chatgpt4o or higher

Even Chinese companies products sold to West may choose to go with US based AI services and APIs, nevermind actual US companies like Apple which still make the bulk of their iphones in China but is integrating with OpenAI etc

Have you seen the 2008 batman dark knight movie or the ex machina movie and references to project bluebook? In each case, fiction is now reality, given the demo OpenAI showcased with chathpt4o real time vision and audio multimodal capabilities and now soon that being baked into all Apple devices at OS layer, this solves the quality data problem, as every phone will be collecting data for US AI.... its more than to help blind man cross street as in the demo, forget for a moment the CIA/NSA aspect of all of this, but soon US will have more data for AI than they know what to do with

All Im saying is this is USA first opening salvo in the upcoming AI Wars and China shouldnt make light of it...AI is the ultimate frontier and the first to get to and scale up AGI and hyperintelligence stays on top forever

 

[tphuang]

Intelligence (AI layers) will be the ultimate value add... China can make the physical products but the AI layer will be running US AI... just like Oppo, etc are Chinese branded phones but run US based Google andriod .... facebook rayban glasses are also made in China, but run meta AI as its brains

That's simply not true, since all the flagship phones have their own separate AI layer. You should look up Xiaomi's Hyper OS and how it interacts with AOSP

The Robosen Optimus Prime comes to mind, its an awesome transformer toy for $999 , made in China but by a US company... if they integrate it later with vision, audio, LLM, etc it will almost certainly be with chatgpt4o or higher

That's one company. There are many Chinese companies.
Whose LLM do you think Ubtech use?
Whose auto LLMs do you think Chinese automakers use

Even Chinese companies products sold to West may choose to go with US based AI services and APIs, nevermind actual US companies like Apple which still make the bulk of their iphones in China but is integrating with OpenAI etc

Do you have any example or prove of this?

Have you seen the 2008 batman dark knight movie or the ex machina movie and references to project bluebook? In each case, fiction is now reality, given the demo OpenAI showcased with chathpt4o real time vision and audio multimodal capabilities and now soon that being baked into all Apple devices at OS layer, this solves the quality data problem, as every phone will be collecting data for US AI.... its more than to help blind man cross street as in the demo, forget for a moment the CIA/NSA aspect of all of this, but soon US will have more data for AI than they know what to do with

All Im saying is this is USA first opening salvo in the upcoming AI Wars and China shouldnt make light of it...AI is the ultimate frontier and the first to get to and scale up AGI and hyperintelligence stays on top forever

All you've done is working yourself up for no reason, because you can only see how LLM is employed in America.

I'm not convinced OpenAI will have the most important LLM in America longer term. Meta and Llama are producing more useful models that can be utilized by edge devices and many startups.

 

[ougoah]

But we gotta admit that AGI is the holy grail technology goal. Whoever gets to real AGI first and is able to commercialise it fastest wins isn't an exaggeration.

 

[Overbom]

But we gotta admit that AGI is the holy grail technology goal. Whoever gets to real AGI first and is able to commercialise it fastest wins isn't an exaggeration.

Reaching AGI probably means reaching ASI shortly after which tbh would make the whole "I reached AGI first" race irrelevant.

 

[tamsen_ikard]

But we gotta admit that AGI is the holy grail technology goal. Whoever gets to real AGI first and is able to commercialise it fastest wins isn't an exaggeration.

There is no winner in technology development. No technology has ever been restricted to one single country when every country on the planet was hell bent on having it. Yes, there have been technology gaps in the past, but that's because no one took a certain technology seriously, or they lacked the necessary wealth, or their society was busy with something else like getting independent from colonialism, or developing basic industry like steel.

There is no magic in AI. Its just software. And like all software, all you need is programming. And like all software, you can copy it, steal it, do whatever you want with it. There is nothing in this world, that can keep a software secret when national security is involved.

Whatever AI tech is developed by anyone, regardless of how awesome it is, it will be copied, or derived.

BTW, GPU is also not a bottleneck, cause again, if model is the key here, just copy the model. Moreover, Just replace 1 super good GPU with 10 bad GPU. All GPU does is solve massively parallel problems. So, there is no harm in making it even more parallel.

 

[tphuang]

btw, I totally agree with this post

https://twitter.com/i/web/status/1800722313840836750

LLMs and AI computation resources will be commoditized. AI computation resources already is commoditized in China as we can see with everyone lowering prices recently.

 

[9dashline]

btw, I totally agree with this post

https://twitter.com/i/web/status/1800722313840836750

LLMs and AI computation resources will be commoditized. AI computation resources already is commoditized in China as we can see with everyone lowering prices recently.

Cutting edge has moved beyond just LLM

Native multimodality plus multi architecture (such as diffusion plus transformer) is the new thing that makes Kling and Sora and chatgpt4o possible

Its now about vision, audio and ability to process language all in one nueral network model... put a small version of this into a robotic humaniod (unitree) or leverage wifi6/5g for internet cloud API access and suddenly you have a T800

Between this and autonomous drones, the future of warfare will be largely unmanned

And the whole demographic problem is no longer a problem

Entire swabs of jobs, both intellectual and manual labor will disappear overnight... when humanoids cost less than KIA cars then everyone who can afford a Tesla or Apple VR will also be getting themselves an humaniod irobot

As for the rest, the high level decisions a manager, supervisor, director or VP or CEO makes will be reduced to the electricity cost of running AI inference models.... aka pennies .... most intellectual labor will go to zero cost with AI pushing humans out entirely

I dont think USG would have given its tech giants green light to go all in on AI if it wasnt for the existential pressures from China in terms of Great Power Competition

 

[tphuang]

Cutting edge has moved beyond just LLM

Native multimodality plus multi architecture (such as diffusion plus transformer) is the new thing that makes Kling and Sora and chatgpt4o possible

Its now about vision, audio and ability to process language all in one nueral network model... put a small version of this into a robotic humaniod (unitree) or leverage wifi6/5g for internet cloud API access and suddenly you have a T800

Between this and autonomous drones, the future of warfare will be largely unmanned

And the whole demographic problem is no longer a problem

and that will get commoditized also. All the Chinese automakers, smart phone OEMs have their own large models. Most of them incorporate what large AI companies provide and some of them having their own models.

Either way, the key is the data. The only thing OpenAI might be getting additional now is iPhone (outside of China)'s data. This is nothing compared to all the driving data you need for ADAS or manufacturing data you need for smart factories or industrial data you need for smart industries or robot interaction data you need for AI robots. Nobody is going to sell that to OpenAI

 

[9dashline]

and that will get commoditized also. All the Chinese automakers, smart phone OEMs have their own large models. Most of them incorporate what large AI companies provide and some of them having their own models.

Either way, the key is the data. The only thing OpenAI might be getting additional now is iPhone (outside of China)'s data. This is nothing compared to all the driving data you need for ADAS or manufacturing data you need for smart factories or industrial data you need for smart industries or robot interaction data you need for AI robots. Nobody is going to sell that to OpenAI

They can simulate it, see nvidia Issac Sim... even unitree g1 robot used Unreal Engine to do simulation for Reinforcement Learning etc....

Years ago back in 2017, OpenAI had a DeepDrive.io that open sourced a bot that learned to drive (trained) inside of GTA, and then drove very well in all of GTA roads, they removed it from github after rockstar threatened a lawsuit

Eventually "virtualization" as in simulated data will have to be the way.... to keep scaling etc.... Google Deepmind said back in 2017 that with AlphaGoZero, it didnt train on any human games and bootstrapped entirely from zero, taught itself tabula rosa etc