News on China's scientific and technological development.

tacoburger said:

NGL the progress of the A.I industry and China's response to it has left me jaded and realize that there's definitely an issue with Chinese innovation. Like China is still behind America in many areas, space, semiconductors, engines etc etc. But that's fine, China is catching up, there's no way such a poor nation could complete with America, it's already been amazing the progress China has made in this mature fields when 30 years ago the country was struggling to feed itself, those are the usual excuses.

But A.I? China was supposed to be neck and neck with America in that area. Even western institutions and mainstream media admits that, china is equal or even surpassing America in A.I. There's so many important A.I research papers with chinese authors, so many money poured into computing/data centers, tech companies and A.I labs. So why is it that every important A.I model released in the last 5 years has been American? Aplhafold, GTP series and it's LLM clones by google/microsoft, video generation models, deepfakes, stable diffusion and other A.I art models etc etc. Sure china catches up within a matter of months, and sometimes releases better versions, but why can't they be first and capture the market? It's not helping their image when they're always on the backfoot, it very much looks like copying. They clearly have the talent, expertise and manpower, LLM were greatly improved with some chinese developed transformer architecture and ChatGTP's development involved some chinese scientists.

Just look at the response to ChatGTP. The entire chinese A.I industry was clearly caught off guard by the success and power of it. Dozens of companies and research labs have announced that they are starting programs to create their own ChatGTP version in the last 5 months. Companies that have already worked on LLMs like Baidu with ERNIE are rushing to improve it and integrate it with their digital ecosystem. But why? LLMs aren't new and GTP-3 was already making big waves back in 2020.

Even an average joe like me could see that GTP-3 was a game-changer that had big potential. You would think that the chinese tech and A.I companies on the forefront of A.I research and development would have seen the potential of GTP-3 and started R&D of similar LLM years ago. But nope, it's clear that most of the companies trying to develop a LLM only just started development only after the success of ChatGTP. The room was already on fire since GTP-3, but they had to wait until they themselves caught fire with ChatGTP and now GTP-4 before running for the fire extinguisher . Even companies that were working LLMs before ChatGTP came out like Baidu admitted that their LLMs are nowhere close to ChatGTP or GTP-4, but even then, why did Baidu not even think of integrating even an average LLM into their digital ecosystem? Why wait until the open AI and microsoft deal to integrate chatGTP into Bing and other microsoft services before Baidu suddenly had the same idea?

It's honestly pathetic to see. What are all those chinese tech companies and A.I labs doing? Were they in some alternate universe when GTP-3 was released back in 2020? Did they not think that GTP-3 wasn't going to be a big deal, or that GTP-4 wasn't going to be a major improvement? Judging by their development speed, it wouldn't take long to catch up, but it's clear that if some aspects of the A.I industry and development were different, China wouldn't have to play catch up at all and would be the ones driving the industry forwards instead of the other way around. It's so frustrating, they clearly have the money, talent, computing power and technical knowledge but something is clearly holding them back. I don't know if it's leadership issues, culture, management issues or censorship.

And like I said, it's not just LLMs. Does China have an answer to AlphaFold? Dalle-2 kickstarted the wave of A.I we're all seeing.

It this always going to be the trend? America releases some brand new novel A.I development, like a self driving A.I that can drive better than any human, or a new A.I architecture that needs a hundred times less data and computing power to train while China always scrambles to play catch up and always be a few months to a year behind?

I hope that the success of ChatGTP and GTP-4 makes the chinese A.I sector take a good hard look at itself and what changes it can make to innovate better. Or actually release their products instead of letting it rot inside of a sever farm.

Click to expand...

You can freak out once Chat-GPT is actually being used to do things that are materially useful rather than just spitting out impressive human BS mimicry.

People here need to stop basing their assessments of technological progress and prowess around spectacle and hype. There are forms of AI that meaningfully matter for real innovation, like self driving AI, AlphaFold, AI used to generate aerodynamic designs, etc. but Chat-GPT is mostly just a giant fault prone index with a human intuitive interaction interface. Literally its most impressive feat is that it's good at interpreting and mimicking human text patterns. That's it. What makes lay persons go wow isn't what always what makes the most significant material impacts. There are thousands of agricultural innovations being explored right now that probably matter more than Chat-GPT.

tacoburger said:

NGL the progress of the A.I industry and China's response to it has left me jaded and realize that there's definitely an issue with Chinese innovation. Like China is still behind America in many areas, space, semiconductors, engines etc etc. But that's fine, China is catching up, there's no way such a poor nation could complete with America, it's already been amazing the progress China has made in this mature fields when 30 years ago the country was struggling to feed itself, those are the usual excuses.

But A.I? China was supposed to be neck and neck with America in that area. Even western institutions and mainstream media admits that, china is equal or even surpassing America in A.I. There's so many important A.I research papers with chinese authors, so many money poured into computing/data centers, tech companies and A.I labs. So why is it that every important A.I model released in the last 5 years has been American? Aplhafold, GTP series and it's LLM clones by google/microsoft, video generation models, deepfakes, stable diffusion and other A.I art models etc etc. Sure china catches up within a matter of months, and sometimes releases better versions, but why can't they be first and capture the market? It's not helping their image when they're always on the backfoot, it very much looks like copying. They clearly have the talent, expertise and manpower, LLM were greatly improved with some chinese developed transformer architecture and ChatGTP's development involved some chinese scientists.

Just look at the response to ChatGTP. The entire chinese A.I industry was clearly caught off guard by the success and power of it. Dozens of companies and research labs have announced that they are starting programs to create their own ChatGTP version in the last 5 months. Companies that have already worked on LLMs like Baidu with ERNIE are rushing to improve it and integrate it with their digital ecosystem. But why? LLMs aren't new and GTP-3 was already making big waves back in 2020.

Even an average joe like me could see that GTP-3 was a game-changer that had big potential. You would think that the chinese tech and A.I companies on the forefront of A.I research and development would have seen the potential of GTP-3 and started R&D of similar LLM years ago. But nope, it's clear that most of the companies trying to develop a LLM only just started development only after the success of ChatGTP. The room was already on fire since GTP-3, but they had to wait until they themselves caught fire with ChatGTP and now GTP-4 before running for the fire extinguisher . Even companies that were working LLMs before ChatGTP came out like Baidu admitted that their LLMs are nowhere close to ChatGTP or GTP-4, but even then, why did Baidu not even think of integrating even an average LLM into their digital ecosystem? Why wait until the open AI and microsoft deal to integrate chatGTP into Bing and other microsoft services before Baidu suddenly had the same idea?

It's honestly pathetic to see. What are all those chinese tech companies and A.I labs doing? Were they in some alternate universe when GTP-3 was released back in 2020? Did they not think that GTP-3 wasn't going to be a big deal, or that GTP-4 wasn't going to be a major improvement? Judging by their development speed, it wouldn't take long to catch up, but it's clear that if some aspects of the A.I industry and development were different, China wouldn't have to play catch up at all and would be the ones driving the industry forwards instead of the other way around. It's so frustrating, they clearly have the money, talent, computing power and technical knowledge but something is clearly holding them back. I don't know if it's leadership issues, culture, management issues or censorship.

And like I said, it's not just LLMs. Does China have an answer to AlphaFold? Dalle-2 kickstarted the wave of A.I we're all seeing.

It this always going to be the trend? America releases some brand new novel A.I development, like a self driving A.I that can drive better than any human, or a new A.I architecture that needs a hundred times less data and computing power to train while China always scrambles to play catch up and always be a few months to a year behind?

I hope that the success of ChatGTP and GTP-4 makes the chinese A.I sector take a good hard look at itself and what changes it can make to innovate better. Or actually release their products instead of letting it rot inside of a sever farm.

Click to expand...

Would your rant and especially this sentence " there's definitely an issue with Chinese innovation" make sense if you replaced Chinese with "German", "Japanese", or "South Korean ?" I haven't seen any LLMs from those countries.

Again, ,just because you don't know what's going on does not mean that things are not happening .


April 29, 2021

latenlazy said:

You can freak out once Chat-GPT is actually being used to do things that are materially useful rather than just spitting out impressive human BS mimicry.

Click to expand...

That was what people said about GTP 3. And those impressive human BS mimicry consist of shit like being able to automate tons of service jobs or enhance digital ecosystems. Being able to spit out a book on command isn't impressive? Billions were already being spent on large data automation, GTP can do so with ease and spin a fancy yarn at the same time.

Next thing you're gonna tell me is that a fully humannoid robot isn't impressive because it's just a mimicry of a human form. Aircraft aren't impressive because they just mimic birds. Not to mention that ChatGTP is still very new, there's tons of potential uses still being figured out, people are trying to intergrate it into games, search engines, using it for propaganda, writing movie scripts, coding etc etc. Not to mention integrating it with other media A.I models like deepfakes, video/image generation A.I.

latenlazy said:

There are forms of AI that meaningfully matter for real innovation, like self driving AI, AlphaFold, AI used to generate aerodynamic designs, etc.

Click to expand...

It's not like China is making big waves there too. Did china come up with AlphaFold? Is there any chinese developed A.I models being used to model aerodynamic designs?

latenlazy said:

Literally its most impressive feat is that it's good at interpreting and mimicking human text patterns. That's it.

Click to expand...

And that isn't impressive? Believe it or not, language is one of the most important tools of being a human. Language is everything. You can make books, spam billions of people with propaganda, suggest new ideas, automate service jobs, mental health services, a large part of the internet is just people talking to each other etc etc, all with just a good language model. This entire forum is just people talking to each other. If someone uses a LLM to infiltrate every public forum on the internet, talk like a real person and influence billions of people over the internet, are you still going to claim that it's useless?

It's not like improvements aren't being made. If GTP-5 is so advanced that it's able to write code at the levels of a human, would you still say that it's unimpressive? There's already talk of multi-model LLM being paired with voice generation, video/image generation A.I, so you can make video/movies and porn on the as you wish.

latenlazy said:

There are thousands of agricultural innovations being explored right now that probably matter more than Chat-GPT.

Click to expand...

And America is once again at the forefront. China is barely just considering letting chinese farmers grow GMO crops, for livestock use only. And who knows? LLM could be key to AGI, it's not like human brains aren't just really good pattern recognition machines that's really good at predicating stuff.

Topazchen said:

Would your rant and especially this sentence " there's definitely an issue with Chinese innovation" make sense if you replaced Chinese with "German", "Japanese", or "South Korean ?" I haven't seen any LLMs from those countries.

Click to expand...

And is China trying to complete with this countries? I thought China was trying to be a superpower equal to America, not complete with this 2nd rate regional powers. Fine, if you want to China to be on the same level as Japan or germany, go ahead and forgive this incompetency.

Topazchen said:

Again, ,just because you don't know what's going on does not mean that things are not happening .

Click to expand...

It's clear that chinese A.I and tech sector doesn't know what's going on. Again, ChatGTP shocked them, it's clear as day, months after ChatGTP came out, dozens of startups, labs, universities, companies are jumping on the LLM bandwagon to create their own ChatGTP. Not "we started development a few months/years ago and are trying to make a better version of the model we're trying working on" but rather "we only started work on LLM after chatGTP came out". Even Baidu, who had their ERNIE model developed since 2020, admitted to being blindsided by ChatGTP and noted that ERNIE wasn't ready to be used, but they had to rush it out due to pressure from ChatGTP.

This is unacceptable. Even I knew that LLM were going to be a large deal when GTP-3 came out. But a whole bunch of chinese A.I researchers who are meant to be on the forefront of A.I research were sleeping on the job I guess. This is like the space race, but the Americans only became panicked after the soviets send Yuri Gagarin into space and dismissed Sputnik as an useless trinket that didn't do much more than bleep, which is true, but clearly Sputnik was just the first signs of something greater to come. AGI is easily the most important technology of the century, even "useless" developments like people are claiming ChatGTP to be have to be watched and developed like an active nuclear weapons program.

So again, what the hell were they doing? Did they not think that LLM were an important technology? Even after GTP-3? It's not like they were coming out with any new novel A.I models of their own.

tacoburger said:

NGL the progress of the A.I industry and China's response to it has left me jaded and realize that there's definitely an issue with Chinese innovation. Like China is still behind America in many areas, space, semiconductors, engines etc etc. But that's fine, China is catching up, there's no way such a poor nation could complete with America, it's already been amazing the progress China has made in this mature fields when 30 years ago the country was struggling to feed itself, those are the usual excuses.

But A.I? China was supposed to be neck and neck with America in that area. Even western institutions and mainstream media admits that, china is equal or even surpassing America in A.I. There's so many important A.I research papers with chinese authors, so many money poured into computing/data centers, tech companies and A.I labs. So why is it that every important A.I model released in the last 5 years has been American? Aplhafold, GTP series and it's LLM clones by google/microsoft, video generation models, deepfakes, stable diffusion and other A.I art models etc etc. Sure china catches up within a matter of months, and sometimes releases better versions, but why can't they be first and capture the market? It's not helping their image when they're always on the backfoot, it very much looks like copying. They clearly have the talent, expertise and manpower, LLM were greatly improved with some chinese developed transformer architecture and ChatGTP's development involved some chinese scientists.

Just look at the response to ChatGTP. The entire chinese A.I industry was clearly caught off guard by the success and power of it. Dozens of companies and research labs have announced that they are starting programs to create their own ChatGTP version in the last 5 months. Companies that have already worked on LLMs like Baidu with ERNIE are rushing to improve it and integrate it with their digital ecosystem. But why? LLMs aren't new and GTP-3 was already making big waves back in 2020.

Even an average joe like me could see that GTP-3 was a game-changer that had big potential. You would think that the chinese tech and A.I companies on the forefront of A.I research and development would have seen the potential of GTP-3 and started R&D of similar LLM years ago. But nope, it's clear that most of the companies trying to develop a LLM only just started development only after the success of ChatGTP. The room was already on fire since GTP-3, but they had to wait until they themselves caught fire with ChatGTP and now GTP-4 before running for the fire extinguisher . Even companies that were working LLMs before ChatGTP came out like Baidu admitted that their LLMs are nowhere close to ChatGTP or GTP-4, but even then, why did Baidu not even think of integrating even an average LLM into their digital ecosystem? Why wait until the open AI and microsoft deal to integrate chatGTP into Bing and other microsoft services before Baidu suddenly had the same idea?

It's honestly pathetic to see. What are all those chinese tech companies and A.I labs doing? Were they in some alternate universe when GTP-3 was released back in 2020? Did they not think that GTP-3 wasn't going to be a big deal, or that GTP-4 wasn't going to be a major improvement? Judging by their development speed, it wouldn't take long to catch up, but it's clear that if some aspects of the A.I industry and development were different, China wouldn't have to play catch up at all and would be the ones driving the industry forwards instead of the other way around. It's so frustrating, they clearly have the money, talent, computing power and technical knowledge but something is clearly holding them back. I don't know if it's leadership issues, culture, management issues or censorship.

And like I said, it's not just LLMs. Does China have an answer to AlphaFold? Dalle-2 kickstarted the wave of A.I we're all seeing.

It this always going to be the trend? America releases some brand new novel A.I development, like a self driving A.I that can drive better than any human, or a new A.I architecture that needs a hundred times less data and computing power to train while China always scrambles to play catch up and always be a few months to a year behind?

I hope that the success of ChatGTP and GTP-4 makes the chinese A.I sector take a good hard look at itself and what changes it can make to innovate better. Or actually release their products instead of letting it rot inside of a sever farm.

Click to expand...

Get your panties unknotted; you don't sound superior or smarter by lecturing others that they should have done better. Anybody can do that and everybody wants to do better but very few know exactly how. "What are you guys doing? You should have 15 EMALS nuclear supercarriers by now; you should have had an exoscale computer in 2012; Chinese GMO tech should be the best in the world! You guys are pathetic!" Yeah no, unless you've done those things, you're the pathetic one for sounding like an angry child. It's not easy or cheap doing something first so unless you want to join a Chinese company and see if you can put out Chinese AI products faster than the US, you're just being a useless backseat driver like some fat NBA fan howling at Lebron from his couch for missing shots.

Once again, it's very very expensive being the first on the block and it's not particularly useful either. Following closely behind the leader has great benefits in efficiency because you essentially forgo a lot of expensive failures that nobody sees. By allowing someone else to demonstrate a proof of concept before you go all in, you can get essentially the same thing a few months to a year later but at much lowered cost and risk. And in the grand scheme of things, that lag is a very very short time. When China is to overtake the US, it will be because 1. the US is tired and spent from paying a premium all the time trying to get crazy stuff to work (desperately trying to get a jump on China) when it doesn't and 2. China has so much extra reserve resources that it would be holding itself back to stay behind so it must burst forward and start bearing greater costs for innovation. Right now, China's quite comfortable in certain areas staying just a little behind the US cruising at a more efficient energy/speed benefitting from the reduced wind resistance while the US exhausts itself trying to keep a lead that really isn't all that meaningful. And in other areas, it's already ahead of the US and pulling farther apart.

tacoburger said:

Not to mention that ChatGTP is still very new, there's tons of potential uses still being figured out, people are trying to intergrate it into games, search engines, using it for propaganda, writing movie scripts, coding etc etc. Not to mention integrating it with other media A.I models like deepfakes, video/image generation A.I.

Click to expand...

There's a lot of menial intellectual work that ChatGPT can do well, and it's a useful tool in the hands of an expert who can identify the plausible-sounding bullshit it often generates. But there's a reason why models like it will always be limited: they cannot reason. It predicts the next token in a sentence, it has no internal model of the concepts it "reasons" about.

Until that happens, there is no AGI. That's the definition of a general intelligence: a system that can construct conceptual models of information it encounters and and reason using those models. No one has the first clue about how to build anything like that.

tacoburger said:

Did china come up with AlphaFold?

Click to expand...

No, China's biotechnology sector is backward and there's not much funding for this kind of work. That's slowly changing but we're not there yet. Fortunately, China doesn't have to come up with AlphaFold when there's a similar system that's freely available:

tacoburger said:

It's not like improvements aren't being made. If GTP-5 is so advanced that it's able to write code at the levels of a human, would you still say that it's unimpressive? There's already talk of multi-model LLM being paired with voice generation, video/image generation A.I, so you can make video/movies and porn on the as you wish.

Click to expand...

GPT-5, 6, 7,... are just bigger bulldozers. The fact that they're bigger won't change them from bulldozers to AGIs. Yes, they're useful; no, they're not AGI.

tacoburger said:

LLM could be key to AGI, it's not like human brains aren't just really good pattern recognition machines that's really good at predicating stuff.

Click to expand...

No. Nothing being done in AI today is going to lead to AGI. The problem is so difficult that it doesn't even have a definition. If we had the complete connection map of a cockroach's brain, we still couldn't figure out why the cockroach turns left. You think we're going to get machines with human cognition in the lifetime of anyone reading this?

We're at Aristotle's "steam rises and water falls to seek their natural place" level of understanding of this problem.

tacoburger said:

Again, ChatGTP shocked them

Click to expand...

You were there in the room with them? Tell us all about it.

tacoburger said:

AGI is easily the most important technology of the century, even "useless" developments like people are claiming ChatGTP to be have to be watched and developed like an active nuclear weapons program.

Click to expand...

AGI will be the greatest technology in human history. Nobody is anywhere close to it.

Tianjin University developed a new method of designing alloy catalysts for propane dehydrogenation (PDH), a process to convert propane feedstock to propylene. The new catalysts designed by the university use 60% less precious metals than those created from the traditional error-and-trials approach.

Over the years, Tianjin University has created relatively complete patents coverage on the new catalysts for the PDH process.

Paper:

Designing single-site alloy catalysts using a degree-of-isolation descriptor​

Abstract​

Geometrically isolated metal atoms in alloy catalysts can target efficient and selective catalysis. However, the geometric and electronic disturbance between the active atom and its neighbouring atoms, that is, diverse microenvironments, makes the active site ambiguous. Herein, we demonstrate a methodology to describe the microenvironment and determine the effectiveness of active sites in single-site alloys. A simple descriptor, degree-of-isolation, is proposed, considering both electronic regulation and geometric modulation within a PtM ensemble (M = transition metal). The catalytic performance of PtM single-site alloy is examined thoroughly using this descriptor for an industrially important reaction, propane dehydrogenation. The volcano-shaped isolation–selectivity plot reveals a Sabatier-type principle for designing selective single-site alloys. Specifically, for a single-site alloy with a high degree-of-isolation, alternation of the active centre has a great impact on tuning selectivity, validated by the outstanding consistency between experimental propylene selectivity and the computational descriptor.

News:

中国青年报：天津大学取得烷烃脱氢合金催化剂“一键筛选”重要进展​

中国青年报客户端讯（中青报·中青网记者胡春艳 通讯员焦德芳）日前，天津大学取得了开发重要化工催化材料的新进展。该校新能源化工团队通过合金催化剂“孤立度”描述符的构建，成功打破传统实验“试错法”局限，实现烷烃脱氢催化剂“一键筛选”。该研究近日发表于《自然·纳米技术》。

丙烯是全球产量最高、应用最广的基础化工原料之一，在工业生产中作用巨大，是制造塑料、合成橡胶和合成纤维等“三大合成材料”的基本原料，对纺织、制药、医疗等领域有重要价值。“丙烷脱氢法”是目前市场占有率增长最快、最具前景的丙烯生产技术之一。

供图正如爱迪生在上千种导电材料中选择出钨丝，传统丙烯生产工艺采用实验“试错法”研发催化剂，往往需要数月甚至数年时间，消耗大量人力物力财力。天津大学新能源化工团队提出了“催化微环境预测催化性能”的方法。他们创新算法，借助计算机程序发现了烷烃脱氢“单位点”合金催化剂的设计新途径。基于该方法，研发新型铂基催化剂的贵金属用量降低超60%，可大幅度节约生产成本，提升了催化剂市场竞争力。

据介绍，近年来，天津大学新能源化工团队探索形成了“催化剂理性设计—精准构筑—应用引领”的研究范例，建立了相对完整的具有自主知识产权的新型高效丙烷脱氢催化剂的专利体系。

tacoburger said:

That was what people said about GTP 3. And those impressive human BS mimicry consist of shit like being able to automate tons of service jobs or enhance digital ecosystems. Being able to spit out a book on command isn't impressive? Billions were already being spent on large data automation, GTP can do so with ease and spin a fancy yarn at the same time.

Click to expand...

You should probably go check out a book called Bullshit Jobs. Services that could be automated away by a text based AI aren't adding much real value to an economy or anyone's lives so much as helping rich people keep more profit for very basic make work service functions. Let me know the last time you found real value from interacting with a telemarketer.

tacoburger said:

Next thing you're gonna tell me is that a fully humannoid robot isn't impressive because it's just a mimicry of a human form. Aircraft aren't impressive because they just mimic birds. Not to mention that ChatGTP is still very new, there's tons of potential uses still being figured out, people are trying to intergrate it into games, search engines, using it for propaganda, writing movie scripts, coding etc etc. Not to mention integrating it with other media A.I models like deepfakes, video/image generation A.I.

Click to expand...

A fully humanoid robot can actually...do physical things. I'm sure there are tons of potential ways to use Chat-GPT. There are also tons of ways we use other objects around us for activities that help us pass the time but don't actually make our or others' lives better.

tacoburger said:

It's not like China is making big waves there too. Did china come up with AlphaFold? Is there any chinese developed A.I models being used to model aerodynamic designs?

Click to expand...

For someone who likes to freak out about China's technology capabilities you really should spend a bit more time learning about where the baseline actually is.

tacoburger said:

And that isn't impressive? Believe it or not, language is one of the most important tools of being a human. Language is everything. You can make books, spam billions of people with propaganda, suggest new ideas, automate service jobs, mental health services, a large part of the internet is just people talking to each other etc etc, all with just a good language model. This entire forum is just people talking to each other. If someone uses a LLM to infiltrate every public forum on the internet, talk like a real person and influence billions of people over the internet, are you still going to claim that it's useless?

Click to expand...

Language is logically complex, sure. Not everything that's logically complex also has real world utility. Let me know when language AI is growing your food and flying you across the ocean and making advanced chips.

If you want to automate away people talking to each other be my guest. Given that I think the point of conversation is not to generate walls of text between two avatars but actually connecting and communicating with other humans, I would say that yeah, that sort of function would be pretty useless.

tacoburger said:

It's not like improvements aren't being made. If GTP-5 is so advanced that it's able to write code at the levels of a human, would you still say that it's unimpressive? There's already talk of multi-model LLM being paired with voice generation, video/image generation A.I, so you can make video/movies and porn on the as you wish.

Click to expand...

Until GTP5 is able to check its own logic, which is not how this AI's model works, it will never be able to write fault free code, and it will *definitely* never generate original code for a function that has never been written before. Using GPT5 to write code is essentially a faster way to rip or copy code out of someone else's library. It's more or less just spitting back recombinations of what other people have written before in the expansive dataset this AI was trained on. This is a useful function for programmers because they don't have to spend so much time digging around Github to look for a function that does what they want to build (programmers these days do a *lot* of copying) but it's also not nearly as impressive or revolutionary or intelligent as you're making it out to be.

tacoburger said:

And America is once again at the forefront. China is barely just considering letting chinese farmers grow GMO crops, for livestock use only. And who knows? LLM could be key to AGI, it's not like human brains aren't just really good pattern recognition machines that's really good at predicating stuff.

Click to expand...

Not every agricultural innovation is GMO related

If you think LLM is the key to AGI then you *really* have no clue how this or other AI works.

manqiangrexue said:

you don't sound superior or smarter by lecturing others that they should have done better. Anybody can do that and everybody wants to do better but very few know exactly how.

Click to expand...

The first part of fixing a problem is acknowledging the that you have an issue. And yes, I do feel a little smarter by thinking that GTP-3 was going to be a games changer, when chinese researchers were sleeping on it until ChatGTP slapped them in the face.

manqiangrexue said:

Once again, it's very very expensive being the first on the block and it's not particularly useful either.

Click to expand...

One big issue. ChatGTP wasn't the first on the block. There was GTP-2 2018, GTP-3 2020, PaLM, LaMDA, flamingo, openAI codex, Megatron, Github copilot and other offshoots like text to image A.I. All released years before ChatGTP. But it wasn't until the immense mainstream success of ChatGTP that everyone in China began to take notice and actually started to hop on the LLM train. Why? Were they lazy? Did they think that LLM was a dead end? Did they not keep up with the advances in their own field? I find it supcisous that the first LLM to really go mainstream was the first one to catch the chinese A.I industry off guard.

A few companies like Huawei or Baidu were working on LLM, but they never released it to the public, or actually made use of it. And it's nothing compared to the dozens of companies currently working on ChatGTP clones in the last 5 months, or the interest in chinese companies asking for a ChatGTP clone for use in their business. Why did Baidu only sought to integrate ERNIE into their search engine after mircosoft announced that they would be integrating GTP-4 into Bing?

manqiangrexue said:

And in the grand scheme of things, that lag is a very very short time.

Click to expand...

When A.I is concerned, a lag time of a few months might as well be a few years. Whoever gets AGI first leapfrogs a century ahead in a matter of months. This is like being WWII, and not caring about who develop nukes first. Did Japan think that America was going to wait a few months to let them develop nukes before they got nuked?

Also, when development of A.I is based on data, being the first, locking in consumers and letting them generate all that sweet sweet data by using your A.I for you is only going to accelerate your lead.

ZeEa5KPul said:

You were there in the room with them? Tell us all about it.

Click to expand...

Surely the dozens of startups, companies, A.I labs and universities that want to create a ChatGTP style startup months after ChatGTP launched isn't a clue? Or the fact that most of those are starting from scratch, instead of accelerating an already existing project, meaning that most of them weren't even working on LLM before ChatGTP went so mainstream that they couldn't afford to ignore it. Baidu CEO outright stated that the public launch of ERNIE was due to pressure from ChatGTP, even having to release it in an incomplete state.

If they were aware, most of them would have started projects of their own before ChatGTP launched, it wasn't like GTP-2 or the myriad of other LLMs weren't a sign of things to come back in 2018.

ZeEa5KPul said:

No one has the first clue about how to build anything like that.

Click to expand...

That depends. Some experts do think that LLM can reach a certain level of understanding with enough parameters. After all, LLM spontaneously already develop some level of reasoning capabilities with enough parameters. There's clearly something going on in the model then just predictive text. We haven't seen what this LLM can do when they reach trillions of parameters. But even if there needs some novel new method I sure as hell expect America to do so first if the sorry state of chinese A.I rearchers can't even hop on the LLM bandwagon until it's slapping them in the face.

latenlazy said:

Let me know the last time you found real value from interacting with a telemarketer.

Click to expand...

Media isn't useless. It's soft power. People laugh at anime and hollywood until you realize that more people are learning Japanese than Mandarin purely for understanding anime and manga.

latenlazy said:

A fully humanoid robot can actually...do physical things.

Click to expand...

And language can make humans do physical things. Like I said, one very scary use. Unleash ChatGTP on every public forum to influence the narrative, shill products and push propaganda. Instead of bots that you can clearly tell apart, it's speaking better than most humans.

latenlazy said:

Let me know when language AI is growing your food and flying you across the ocean and making advanced chips.

Click to expand...

And again, my issue is that China is been ignoring such an obvious and powerful technology. It's worse than being behind. It's not even realizing that that particular area is even worth investing in the first place, or that you're even behind. Why didn't we see such an explosion of activity from the A.I sector in the last few months back in 2020 when GTP-3 was making such buzz? Forgetting developing ChatGTP first, I would have liked them to have saw the potential of GTP-2 in 2018 and started a industry wide effort back then, instead of scrambling now

What's next? America actually figuring out a A.I model that can reach AGI and China ignores it until America is a century ahead?

latenlazy said:

If you think LLM is the key to AGI then you *really* have no clue how this AI works.

Click to expand...

Some experts do think it is. After all, LLM spontaneously develop some level of reasoning capabilities with enough parameters. There's clearly something going on in the model then just predictive text.

latenlazy said:

but it's also not nearly as impressive or revolutionary as you're making it out to be.

Click to expand...

Early days and all that. 2 years ago, I never would have thought that A.I could progress beyond drawing stick figures too. And going "this isn't as impressive as you think is it" isn't helping when my issue is why China didn't develop it earlier or were the first to develop it.

The focus should be less on whether ChatGPT is the path to Artificial General Intelligence, and more on why the West was the first to market on major disruptive technologies. Look, I understand pouring cold water on the over hype around ChatGPT, but similar to the case of block chain technology, it is legitimate to ask why China is not the one making these "game changing" advancements that create entire industries over night. Even if you disagree, fundamentally, that these technologies are "game changing," you can't deny that they've been disruptive, and that China's role has been following / reacting, rather than leading.

Indeed, stuff like this has been the West's traditional source for supremacy, in the first place. Europe's not rich in resources, nor in people - for most of history they've been about 10-15% of humanity and about the same in wealth. Yet, the Scientific Revolution in Europe, the Industrial Revolution in Britain, and all the advances subsequent to them were what enabled Europeans to kick off the Age of Empires, during which they took over 80-90% of the world. It's what created the modern world in which Western hegemony is still present, even if in decline.

So while individually, we might laugh off trends like cryptocurrency as a scam and make jokes about LLMs, when taken as a whole, it still should serve as a reminder that the West remains the source of most disruptive innovations around the world, with only East Asian countries being even close to challenging them on it. It is then legitimate to ask "why". Because honestly, the world can't afford another Western Age of Empires. With how destructive weapons are today, and the speed of technological change, giving the West any disruptive lead on technology is dangerous. Or put it another way - how do we ensure that Western hegemony never happens again, if the West is allowed to keep its lead on disruptive innovation?

tacoburger said:

That was what people said about GTP 3. And those impressive human BS mimicry consist of shit like being able to automate tons of service jobs or enhance digital ecosystems. Being able to spit out a book on command isn't impressive? Billions were already being spent on large data automation, GTP can do so with ease and spin a fancy yarn at the same time.

Next thing you're gonna tell me is that a fully humannoid robot isn't impressive because it's just a mimicry of a human form. Aircraft aren't impressive because they just mimic birds. Not to mention that ChatGTP is still very new, there's tons of potential uses still being figured out, people are trying to intergrate it into games, search engines, using it for propaganda, writing movie scripts, coding etc etc. Not to mention integrating it with other media A.I models like deepfakes, video/image generation A.I.


It's not like China is making big waves there too. Did china come up with AlphaFold? Is there any chinese developed A.I models being used to model aerodynamic designs?

And that isn't impressive? Believe it or not, language is one of the most important tools of being a human. Language is everything. You can make books, spam billions of people with propaganda, suggest new ideas, automate service jobs, mental health services, a large part of the internet is just people talking to each other etc etc, all with just a good language model. This entire forum is just people talking to each other. If someone uses a LLM to infiltrate every public forum on the internet, talk like a real person and influence billions of people over the internet, are you still going to claim that it's useless?

It's not like improvements aren't being made. If GTP-5 is so advanced that it's able to write code at the levels of a human, would you still say that it's unimpressive? There's already talk of multi-model LLM being paired with voice generation, video/image generation A.I, so you can make video/movies and porn on the as you wish.

And America is once again at the forefront. China is barely just considering letting chinese farmers grow GMO crops, for livestock use only. And who knows? LLM could be key to AGI, it's not like human brains aren't just really good pattern recognition machines that's really good at predicating stuff.

And is China trying to complete with this countries? I thought China was trying to be a superpower equal to America, not complete with this 2nd rate regional powers. Fine, if you want to China to be on the same level as Japan or germany, go ahead and forgive this incompetency.

It's clear that chinese A.I and tech sector doesn't know what's going on. Again, ChatGTP shocked them, it's clear as day, months after ChatGTP came out, dozens of startups, labs, universities, companies are jumping on the LLM bandwagon to create their own ChatGTP. Not "we started development a few months/years ago and are trying to make a better version of the model we're trying working on" but rather "we only started work on LLM after chatGTP came out". Even Baidu, who had their ERNIE model developed since 2020, admitted to being blindsided by ChatGTP and noted that ERNIE wasn't ready to be used, but they had to rush it out due to pressure from ChatGTP.

This is unacceptable. Even I knew that LLM were going to be a large deal when GTP-3 came out. But a whole bunch of chinese A.I researchers who are meant to be on the forefront of A.I research were sleeping on the job I guess. This is like the space race, but the Americans only became panicked after the soviets send Yuri Gagarin into space and dismissed Sputnik as an useless trinket that didn't do much more than bleep, which is true, but clearly Sputnik was just the first signs of something greater to come. AGI is easily the most important technology of the century, even "useless" developments like people are claiming ChatGTP to be have to be watched and developed like an active nuclear weapons program.

So again, what the hell were they doing? Did they not think that LLM were an important technology? Even after GTP-3? It's not like they were coming out with any new novel A.I models of their own.

Click to expand...

What's your background in terms of profession if you don't mind me asking? and what's up with your doom and gloom spiel that's really getting tiresome because you're one China is behind/crisis away from becoming the next anti-China zealot with the way you have been lamenting on this perceived backwardness or slow progress of A.I. in China. What's next? China should have been a democracy so that they would have been super duper ahead of the curve against the U.S? I wasn't aware that this very important technological frontier contest/supremacy in A.I. is already over and the U.S. ranked as the top is now firmly cemented. Weak minded person will always be the most alarmist of us all, and you're penchant for dramatic declarations showcase this pathetic sniveling weakness.