Artificial Intelligence thread

[ougoah]

I can tell you right now as we speak, in the company where I work for, one person's job position has been eliminated by LLM. Was doing document processing.

For the edge cases where something goes wrong, it is being handled by another existing office employee

That's not "many years if not decades away from". That's now

I think LLMs are going to be replacing those sorts of roles as they are on balance going to be the better financial decision for corporations. It doesn't mean they are able to replace people in all specialised professions. Okay it'll gradually be consolidating teams and trimming them as they become a more widespread tool used to assist and speed up processes.

The discussion on general intelligence, well LLMs are not general intelligence and maybe never will be. At the moment, an LLM wouldn't be able to perform 1% of my job and to squeeze out that 1% would require some effort unless the process becomes streamlined for the role somehow. I want to imagine this is true for countless professions out there. In theory an LLM may be able to do some roles here and there but it's going to require a lot of checking or refining over time as it trains. The thing is it just gets so many things hilariously wrong and where I'm working they've spent some coin investigating and developing the tech to take over roles with hilariously poor results.

LLMs can obviously be useful but it is not the intelligent or the hyped up world changer many voices have proclaimed it to be. It is not able to navigate any nuance or complexity. In specialised fields, it currently does not have the basic information perhaps and it's a matter of time to develop them to a point where they can perform lots of grunt work. It may be years or decades to get to that level. Obviously software engineers are not going to be learning accounting or geophysics to produce LLMs specifically designed to assist those professions. Therefore I wonder how much people progressing AI are mindlessly investing in the belief that general AI will magically appear and solve those problems by creating a truly infallible piece of digital intelligence that can navigate everything all on its own. It's just that with LLMs, they don't actually really learn. They don't actually think. They just have clever pattern recognition if I understand it at all (admittedly I have no idea on how it functions in entirety). All I know is that if it was all it's claimed to be, they wouldn't be this ridiculously imperfect. Stringing together barely coherent sentences to produce an illusion of intelligible interaction is the only thing it does. This field just happens to overlap a tiny pathetic bit with some ability to write code (not that well) and do a few peripheral tricks by happy coincidence.

Even in shining examples where it has replaced some roles in limited capacity, I think it has already hit its current limits. Even graphic design teams have not suffered massive redundancies around the world. Law clerks are still being employed and these are the bread and butter of what LLMs do best so far.

 

[Overbom]

I think LLMs are going to be replacing those sorts of roles as they are on balance going to be the better financial decision for corporations. It doesn't mean they are able to replace people in all specialised professions. Okay it'll gradually be consolidating teams and trimming them as they become a more widespread tool used to assist and speed up processes.

The discussion on general intelligence, well LLMs are not general intelligence and maybe never will be. At the moment, an LLM wouldn't be able to perform 1% of my job and to squeeze out that 1% would require some effort unless the process becomes streamlined for the role somehow. I want to imagine this is true for countless professions out there. In theory an LLM may be able to do some roles here and there but it's going to require a lot of checking or refining over time as it trains. The thing is it just gets so many things hilariously wrong and where I'm working they've spent some coin investigating and developing the tech to take over roles with hilariously poor results.

LLMs can obviously be useful but it is not the intelligent or the hyped up world changer many voices have proclaimed it to be. Even in shining examples where it has replaced some roles in limited capacity, I think it has already hit its current limits. Even graphic design teams have not suffered massive redundancies around the world. Law clerks are still being employed and these are the bread and butter of what LLMs do best so far.

I already gave an example of what type of job can be heavily automated or even entirely eliminated.

If predictions are proven correct and GPT 5 carries a similar improvement akin from GPT3 to GPT4, but with a heavy focus on agentic behaviour, then even more jobs will be in danger.

For now I see these categories of jobs that are already getting intruded by AI:

Document processing
Customer support
Recruitment/HR
Marketing/Advertising
Graphic Design


Of course this is not applied for technical complex jobs

 

[ougoah]

I already gave an example of what type of job can be heavily automated or even entirely eliminated.

If predictions are proven correct and GPT 5 carries a similar improvement akin from GPT3 to GPT4, but with a heavy focus on agentic behaviour, then even more jobs will be in danger.

For now I see these categories of jobs that are already getting intruded by AI:

Document processing
Customer support
Recruitment/HR
Marketing/Advertising
Graphic Design


Of course this is not applied for technical complex jobs

I agree that some jobs are in danger of being replaced. I never actually mentioned that not being the case. I was specifically referring to many professional jobs (mine included) that are currently not even remotely close to being challenged by LLMs when I made the reply to your post on white collar jobs requiring intelligence (lol btw yes I agree many don't and most don't require much of it).

As for the other side of the discussion on LLMs. It is no doubt useful. It just isn't intelligent. That does not preclude it from being able to fill roles more effectively (and far cheaper faster etc). Being able to replace humans in certain jobs isn't my metric of intelligent or a general AI. LLMs to me is a inferior mirror of humanity. A true general intelligence is a creator not a mimicking program feeding on existing "knowledge" and past data which is what a LLM is. This again does not take away the ability of LLM filling previously human roles.

I think China's major AI projects (many of them LLMs) all understand that LLM are useful and for this reason alone perhaps is why they are being created as well. They probably also understand that LLMs are not the path to general intelligence.

I suspect many people in that field realise this and know it deep down. They just want something to do, they want to be paid handsomely while they do it and LLMs is the lowest hanging fruit and possibly presents the most prominent and well known potential path to general intelligence.

Since no other obvious paths exist currently, people are happy to proclaim this it is.

A LLM is NOT able to advance mathematics or physics but it is able to beat chess players and Go players and so on. That's what it's good at, crunching and processing and recognising patterns. It cannot design a better lunar lander. It cannot hypothesise alternative theories to creation and postulate ways of investigating them. It can create the illusion of composing music and poetry because that's again exactly how the machine works. People confuse these "creativity" for creating. It may be similar for humans indeed but humans have been able to push forward into the unknown throughout history. We went from grunting to making atomic scale calculators with light.

If it were up to a LLM to push the boundary, we'd just have a symphony of inferior grunting. LLM has always been and possibly will always remain an inferior reflection of ourselves and what we're capable of. We teach it everything, it doesn't train itself just processes through more data than any human can. We give it an incomplete and fallible set of "knowledge" and that's all it has to work with. It probably will never be more than that set of incomplete and imperfect knowledge. Anything outside of its pattern trick mechanism and it tells you it's confused and apologises for not having the answer... a human programmer no doubt had some influence there as well even then.

 

[ficker22]

I already gave an example of what type of job can be heavily automated or even entirely eliminated.

If predictions are proven correct and GPT 5 carries a similar improvement akin from GPT3 to GPT4, but with a heavy focus on agentic behaviour, then even more jobs will be in danger.

For now I see these categories of jobs that are already getting intruded by AI:

Document processing
Customer support
Recruitment/HR
Marketing/Advertising
Graphic Design


Of course this is not applied for technical complex jobs

Only true loss and where use of AI is ambiguous is graphic/art design.

 

[9dashline]

This is the type of posts that I was trying to get rid of from semiconductor thread and is apparently going to end up in AI thread now. I don't even know what to respond to here really because it's so filled with conspiracy theories and self pity

The reason Google orkut, Google Wave, Google Buzz, Google Plus, etc werent able to dethrown facebook as top social network was because facebook had first mover advantage , network effect lockin, and marketshare saturation...

Even if China already had a 3nm EUV tech and etc it wouldnt change the fact that in terms of the stack, the rest of the world is more on Azure than on alibaba, more on iphone than on Huawei or Oppo or etc, and iOS and Andriod make up vast majority of smartphone OS layer, with Google Play and Apple itunes/App store taking up almost all of the app ecosystem.... and of course for desktop computer there is Windows.... (btw Windows 11 coming out with Recall mode ostensibly for AI but its to take automatic screenshots of all users desktop/monitor every few seconds etc)

So from a plstform and stack perspective the US is already entrenched and has the world in its grips/cornered... and one would be mistaken to think it wouldnt take full advantage of that....

The idea is all the nations of the world, all the companies in global international scene will bake in AI at every layer , every stack, every process and service, largely by way of API calls via internet to US AI (openai) running on US cloud (Azure) over US networks on US GPU (Nvidia) with payment US dollars. And with its tenacles embedded into all the products (from intelligent toothbrushes to sunglasses, to cars, microwaves, etc etc) and services, and apps and websites etc, it becomes almost like a pervasive ubiquitous superglue that seeps into everything and everywhere......

This iteration of AI has the potential to dusplace more than 90 percent of current intellectual labor globally worldwide...this makes people everywhere more reliant on AI/tech and by extension more reliant on their governments and monopolistic tech companies, but by positioning itself at the very top of this topical pyramid hiarcherial scheme, the US will hold the entire planet capitive and with all the nations of the world beholden and enslaved by America yet once again.... aka digital OPEC, with America once again holding the chokepoints to this time around what is digital gold and virtual oil, that of aritificial intelligence in the form of openai as a service.... AI is the game changer, the reset of the circuit board, the monkey wrench into the equation or roll of dice so to speak....

AI is the great enabler where by US is basically converting and transmuting the human labor of the entire planet, or at least the eq. monetary value thereof, and capturing it, and directly funneling it up to itself at the very top..... its no longer even a tax, but wholesale control and conversion of every nations wealth, resources, decision making....

I hope China already has a strategy in place to counter this, I dont expect China to play "fair", just like US never played fair with Huawei, Tiktok, DJI, TSMC, EUV, etc etc etc

 

[ougoah]

All that may be true but there are and will be increasing alternatives to the US digital ecosystem if it so decides to abuse its privileges beyond thresholds of tolerance even with US allies.

While all existing tech alternatives are currently Chinese, there will eventually be more players in Europe, Korea, Japan, South Asia and possibly other places especially if there are motivations to move away from a monopolistic parasite behaviour. This ensures the US will not be immediately abusing it like with the rest of the techno-feudalism competition currently playing out between the US and non US aligned world. Europe may not learn its lesson early enough but as long as China present alternatives, there is balance and choice available.

The threat though is falling into the trap with dependency established. In terms of winning over nations into its own ecosystem, China has no hope of doing so. Certainly not with US allies. The only hope is if Chinese alternatives are leaps and bounds superior and that simply isn't the case or on the path of becoming the case with GPUs, networks or integrating ML. Nevermind the cultural issue. Face it, China has zero chance of overcoming the fact that US stands with cultural allies whereas China stands alone. The rest of the Global South is too busy figuring out how they're going to avoid the middle income trap even if it gets to that stage without becoming a full time sepoy. China needs revolutionary tech to surpass the US and allies.

 

[tphuang]

The reason Google orkut, Google Wave, Google Buzz, Google Plus, etc werent able to dethrown facebook as top social network was because facebook had first mover advantage , network effect lockin, and marketshare saturation...

Even if China already had a 3nm EUV tech and etc it wouldnt change the fact that in terms of the stack, the rest of the world is more on Azure than on alibaba, more on iphone than on Huawei or Oppo or etc, and iOS and Andriod make up vast majority of smartphone OS layer, with Google Play and Apple itunes/App store taking up almost all of the app ecosystem.... and of course for desktop computer there is Windows.... (btw Windows 11 coming out with Recall mode ostensibly for AI but its to take automatic screenshots of all users desktop/monitor every few seconds etc)

So from a plstform and stack perspective the US is already entrenched and has the world in its grips/cornered... and one would be mistaken to think it wouldnt take full advantage of that....

The idea is all the nations of the world, all the companies in global international scene will bake in AI at every layer , every stack, every process and service, largely by way of API calls via internet to US AI (openai) running on US cloud (Azure) over US networks on US GPU (Nvidia) with payment US dollars. And with its tenacles embedded into all the products (from intelligent toothbrushes to sunglasses, to cars, microwaves, etc etc) and services, and apps and websites etc, it becomes almost like a pervasive ubiquitous superglue that seeps into everything and everywhere......

This iteration of AI has the potential to dusplace more than 90 percent of current intellectual labor globally worldwide...this makes people everywhere more reliant on AI/tech and by extension more reliant on their governments and monopolistic tech companies, but by positioning itself at the very top of this topical pyramid hiarcherial scheme, the US will hold the entire planet capitive and with all the nations of the world beholden and enslaved by America yet once again.... aka digital OPEC, with America once again holding the chokepoints to this time around what is digital gold and virtual oil, that of aritificial intelligence in the form of openai as a service.... AI is the game changer, the reset of the circuit board, the monkey wrench into the equation or roll of dice so to speak....

AI is the great enabler where by US is basically converting and transmuting the human labor of the entire planet, or at least the eq. monetary value thereof, and capturing it, and directly funneling it up to itself at the very top..... its no longer even a tax, but wholesale control and conversion of every nations wealth, resources, decision making....

I hope China already has a strategy in place to counter this, I dont expect China to play "fair", just like US never played fair with Huawei, Tiktok, DJI, TSMC, EUV, etc etc etc

hmm, you haven't shown any evidence that you need 3nm EUV tech for AI chips, because that doesn't exist.

IoS and Android dominance existed independently of AI. Same with Windows.

US cloud company advantages also existed before AI. There is no evidence AI is helping American company gain market share, because it doesn't exist. We just recently saw Egypt building AI data center fully using Huawei technology.

Most of what you are saying here is just an emotional rant. Honestly, it's quite tiresome.

i will just give an example
"And with its tenacles embedded into all the products (from intelligent toothbrushes to sunglasses, to cars, microwaves, etc etc) and services, and apps and websites etc, it becomes almost like a pervasive ubiquitous superglue that seeps into everything and everywhere......"
this is just factually false here. America is way behind China in incorporation of AI into electronics and appliance products

 

[9dashline]

hmm, you haven't shown any evidence that you need 3nm EUV tech for AI chips, because that doesn't exist.

IoS and Android dominance existed independently of AI. Same with Windows.

US cloud company advantages also existed before AI. There is no evidence AI is helping American company gain market share, because it doesn't exist. We just recently saw Egypt building AI data center fully using Huawei technology.

Most of what you are saying here is just an emotional rant. Honestly, it's quite tiresome.

i will just give an example
"And with its tenacles embedded into all the products (from intelligent toothbrushes to sunglasses, to cars, microwaves, etc etc) and services, and apps and websites etc, it becomes almost like a pervasive ubiquitous superglue that seeps into everything and everywhere......"
this is just factually false here. America is way behind China in incorporation of AI into electronics and appliance products

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.

 

[9dashline]

In fact, very recently, OpenAI just announced that they've opened APIs to their ChatGPT-4.0 as an Omni, where anybody with a ChatGPT account, $20 subscription, can leverage it, and with very little knowledge of coding, can basically script it where they integrate with their personal or corporate OneDrive account, or even Google Drive account, and it's able to search through all of their documents and make sense of it, so that they can basically use a customized version of their GPT for their business to enhance their business directly, without even the knowledge and the effort of so-called fine-tuning a large language model. So, this is where we're at now, and I work in IT, and I see basically everybody using it now. It's almost like the new search engine. Now they're coming out with that desktop model soon, first on Mac, then on Windows, where effectively you're doing a real-time live screen sharing with the AI of shoulder surfing, and contextually seeing everything that you're doing, and you're able to talk to it in real-time to get real-time natural language conversation and feedback, to have it assist and help you with all sorts of tasks. Next, Microsoft's going to bake in this AI with their PowerShell, so that you can basically control your entire computer just on voice command. So, the potential is limitless, and of course, there's this coordinated effort of the likes I've never seen before in my entire life. I've never seen this level of coordinated effort by all the tech companies to really target and spearhead this, and to focus on this on a level that is unprecedented, and that to me feels like they've been given a green light directive from the highest level of the US government to pursue this, to put all the bets on this, and to go all in.

 

[9dashline]

I agree that some jobs are in danger of being replaced. I never actually mentioned that not being the case. I was specifically referring to many professional jobs (mine included) that are currently not even remotely close to being challenged by LLMs when I made the reply to your post on white collar jobs requiring intelligence (lol btw yes I agree many don't and most don't require much of it).

As for the other side of the discussion on LLMs. It is no doubt useful. It just isn't intelligent. That does not preclude it from being able to fill roles more effectively (and far cheaper faster etc). Being able to replace humans in certain jobs isn't my metric of intelligent or a general AI. LLMs to me is a inferior mirror of humanity. A true general intelligence is a creator not a mimicking program feeding on existing "knowledge" and past data which is what a LLM is. This again does not take away the ability of LLM filling previously human roles.

I think China's major AI projects (many of them LLMs) all understand that LLM are useful and for this reason alone perhaps is why they are being created as well. They probably also understand that LLMs are not the path to general intelligence.

I suspect many people in that field realise this and know it deep down. They just want something to do, they want to be paid handsomely while they do it and LLMs is the lowest hanging fruit and possibly presents the most prominent and well known potential path to general intelligence.

Since no other obvious paths exist currently, people are happy to proclaim this it is.

A LLM is NOT able to advance mathematics or physics but it is able to beat chess players and Go players and so on. That's what it's good at, crunching and processing and recognising patterns. It cannot design a better lunar lander. It cannot hypothesise alternative theories to creation and postulate ways of investigating them. It can create the illusion of composing music and poetry because that's again exactly how the machine works. People confuse these "creativity" for creating. It may be similar for humans indeed but humans have been able to push forward into the unknown throughout history. We went from grunting to making atomic scale calculators with light.

If it were up to a LLM to push the boundary, we'd just have a symphony of inferior grunting. LLM has always been and possibly will always remain an inferior reflection of ourselves and what we're capable of. We teach it everything, it doesn't train itself just processes through more data than any human can. We give it an incomplete and fallible set of "knowledge" and that's all it has to work with. It probably will never be more than that set of incomplete and imperfect knowledge. Anything outside of its pattern trick mechanism and it tells you it's confused and apologises for not having the answer... a human programmer no doubt had some influence there as well even then.

So first there was just the diffusion architecture such as the likes of MidJourney and DALL·E2 and Stable Diffusion. Then came the Diffusion Plus Transformer using attention mechanism architecture. And this is where we get KLING and the likes of Sora. And you know, it's getting to the level where it's good enough to be used in production. So we don't have to, like, it used to be, hey, you have to set up a scene, you have to do a 3D Studio Max, you have to play, hey, people, you have to do the modeling. And then once you model it, you have to texture it, you have to make sure, you know, polygons are all good, level of detail, and then use something like V-Ray to do the actual rendering in terms of the 3D realistic physics interaction with light, ray tracing, pass tracing, do all these very complicated, very intensive calculations just per frame. And then to do a short snippet video, you had to render hundreds of thousands of these individual frames, and you needed a render farm to do this. And now with the likes of Sora and KLING, one sentence, you know, prompt can generate something equal or better than, no setting up the scene, no setting up the stage, no doing the render pipeline, no need for complicated physics calculations. The AI basically, quote unquote, generates all of it generatively, which is much cheaper than calculating it brute force in a vector, you know, style using raw light base physics and the such. So this is just one example of the power of the exponential content that AI has afforded us that we no longer need to even use Photoshop and 3D Studio Max and V-Ray and this, that, and the other.