Timestamps (Powered by Merlin AI) 00:03-Generative AI is more significant for the industrial physical world 02:16-AI enables expertise to become more important 06:39-Higher industrialized economies are more optimistic about AI adoption 09:04 - Domain-specific AI agents are crucial for industrial applications 13:14-Capturing and encoding expert knowledge in the knowledge age 15:23 - Domain-specific AI agents are revolutionizing the industrial world. 19:32 - Domain-specific AI agents are goal-oriented with planning and reasoning capabilities. 21:22-Industrials are the early adopters of AI technology 25:30-Recurrent Loop in planning and reasoning is key for future AI agents 27:20-Incorporating recurrence and problem-solving in AI models. 31:14 - Domain-Specific AI Agents enhance problem-solving abilities in industrial settings
Lack of AI people with domain knowledge is a big problem. A lot of them know a lot about AI but nothing about banking, manufacturing, health care, or any part of the real world.
this is very true. AI skills need to be taught to the domain-specific engineers e.g. I have a much better grasp working in healthcare having done biomedical engineering, but I cannot guarantee the same translation to another domain.
@@hifzul123 Within your domain you can be recognized as an SME. The key issue in using your own AI's is loading domain-specific information to the knowledge base. Consider learning to use GPT4ALL. It lets you run AI's on your own PC. You probably want i5 or better because they run faster with more threads. I'm running an i7 with 16 threads. The Llama 3.2 1B model is a good choice for running with your own knowledge.
@@tangobayus Nope. The AI "people" don't need to "have" the knowledge. They scraped the internet and all public libraries and public papers, etc... there is no problem here. Also, it only takes one good source of data. Also, the next gen AI models will create their own data, by interacting with the world. No problems here folks.
And do what with all this AI if many of us are employed? Who can afford the good and services produced by AI? Nice. AI surely can replace CEO/CTO/CFO positions.
Capturing knowledge is challenging. All AI agents should have a CV profile that includes qualifications, work history, experience, and feedback ratings from humans. Then we can interview and onboard the agents using evidence and reference letters.
Feedback ratings from humans have never been reliable. Especially if it’s about something like AI? No chance. A performance mechanism built into blockchain will track its performances per accomplishment (Agentic AI) Fuck humans, humans are bitter and strange, said just as bitterly 😅
I'm going to guess that the answer is actually within your own question. It obviously was the most important part of the presentation and too good to put online here lol.
He said that the person that was helping them train the domain-specific expert level AI was about to retire. Exactly, because anyone who trains domain-specific AI is about to retire their human job, whether they are ready or not. Until these companies have an answer to how they are going to honor the loyalty of these employees with loyalty FROM THE COMPANY that helps them get realistic LONG TERM, sustained training to keep them paid and relevant, the U.S. will not only remain pessimistic about AI, but expect that pessimism to ROCKET upwards as this tech is (pun intended) employed, while humans are no longer, more and more.
You raise an important point about company loyalty during AI transitions. What if companies invested in reskilling programs that help employees become AI operators and domain experts in their field? This could create a win-win where experienced workers guide AI systems while developing new valuable skills
@Custom.AIAgentAcademy Why would they do this? Did they retrain coal miners? Wake up. We are in a TERRIBLE position economically and politically for the integration of AI. Labor union power is at an all time low, mega corporate and oligarch power is at an all time high. The PTBs have been sucking every penny from the bottom to the top; AI and robots are SPECIFICALLY designed to do just that. Without a MAJOR revolution starting almost immediately, we are all F'd.
Would have liked to have heard more about domain-specific AI.. Poe, Causaly (science), Consensus (academic research), and Hebbia (financial services).. and chatboxes.. Character, Claude, Grok, Meta AI, Poe, DeepSeek (China)
You mention some interesting domain-specific AI tools. Have you had experience working with any of them? I'd be curious to hear how they compare in terms of real-world applicability versus general-purpose AI
I hate to be negative but manufacturers have been "capturing" domain specific process data for about 50 years. It is often written down but it is always captured and taught to younger employees. In die casting plants we have 60 process inputs and it is quite complex. I am sure that a chip fab is quite a bit more complex. Capturing the data is not new at all. Tha challenge is to know how to use the data to solve pocess problems, test the results and improve the process. This requires skilled people. AI will not be doing this any time soon.
You are correct and it’s not negative. In particular, in the semiconductor industry, we have long had extensive documentation and processes to capture operational expertise. The exciting opportunity today is that it has become much easier to capture and to operationalize using AI. Because today’s AI can much more readily “speak/understand” natural language and other sources of knowledge from our physical world.
I agree with the first part you said, however LLM does have reason capability but “training” it to consistently generating the right answer is a challenge, at the moment. For domain specific topic, it’s even harder. the training data set is proprietary, low volume, and the LLM was largely trained on noisy data which causing the result to be inconsistent and unreliable. But this can be engineered and corrected over time. It’s no longer impossible.
I disagree! It's not about AI solving the problem. AI as a tool can enable skills development using data and processes in the context of manufacturing.
@@edwardjones856 You have zero reason to say AI won't be doing it soon. How complex is programming? Making entire videos in seconds with a prompt? Imitating voices near perfectly. The AI merely needs to be trained on the correct data sets, the exact same ones that the humans would be using, and given the correct supervision and guidance during training. The presenter just spelled it out in THIS VIDEO. I'm sorry sir, but your insurmountable obstacles are EASILY solved, even at current AI levels and when AGI arrives, AGI will do ALL of the solving: period.
I'm working on a screenwriting project where we are using AI's for feedback. They can improve the writing sometimes but they can't come up with the creative ideas that we provide.
We need AI to increase the leverage of physical workers at the coal face - we need 10x the resources to get off planet and white collar work is obsolete, but smart guys with wrench’s own the future
I think I am not understanding what the speaker wants to say. Even langchain or cot or tot or agent using LLM to create plan Yes, Is it is true that as of 2024 we cannot 100% depend on LLM. Therefore we should go slow in automation. But the reason he is giving is incorrect.
Initially, AI is an enabler. If it emerges in an ethically corrupted milieu it will enable corruption. Alternatively, it could facilitate the emergence of a society that measures performance against commitment, personal agency, responsibility, and equitability, a “great equalizer” promoting objectively wise outcomes. This is an existentially significant choice. Read Amaranthine: How to Create a Regenerative Civilization Using Artificial Intelligence
US wars from the past 30 years have only benefited one nation and it hasn't been the US. Unlikely that the US will end any wars anytime soon since every party is captured except the green.
@@The_Quaalude lol, good luck. All of that tech is proprietary to Taiwan and Taiwan only. Taking that your disabled country has constant leaks of military blueprints to chinese there is doubt that States are able to do anything at all. Nobody is scared or even respects usa rn. And idiots that scream about isolationism just continue to prove that nobody should believe in this failed state
The biggest risk the US has is that almost all the Semiconductor company have their design centres in Israel as well. Now, any electronics have a very high risk of getting compromised, including turning into a bomb as we saw in Lebanon.
I just don’t see how the pace of new knowledge creation does not end once the humans are all laid off or retired. Endless permutation of old by AI is not creation of the new.
Spot on. They are great for regurgitation but the sparks of creativity come from somewhere else. God or whatever you wish to call it. You need a soul to get information from God. AI's don't have one.
I wonder if in the 'expert mode' contexts of GenAI, when we'll stop calling the over-confident, incorrect (and even made-up) responses 'hallucinations' and finally begin to call them 'bullshit' as we would call out GenAI's human analogues. But wait, GenAI has 'PhD level nunchuk skills'. We're talking super-PhD-impossible-to-bullshit level mega brain (with weaponry! ) here. Nah, why question 'it' either - that's far above our mortal pay grades ...
I found this presentation strange. I worked in the 90s creating expert systems. It failed miserably. Why? How we do things today is already inefficient and out of date. Adoption, standards, expertise are all behind R&D and emerging know-how and innovations. Experts are actually the worst at predicting the future and at adapting to the future. Every study done on this topic comes to the same conclusion: experts can't see their own biases. They are slow to change their minds in a rapidly changing world. Agentic AI is important. It is the top layer of AI, but it will grow organically from data and self-reflexion over time. Today's human specialists are tomorrow's dinosaurs.
I too built some expert systems back then. They suffer from combinatorial explosion. I think they have come back in the guise of robotic process automation. Generative AI is not good with executing processes reliably.
Interesting perspective from your 90s expert systems experience. How do you think today's machine learning approaches differ from those earlier expert systems? While experts may have biases, couldn't their practical knowledge still be valuable in training AI to handle real-world edge cases?
![Felix "Unfair" | [Stray Kids : SKZ-PLAYER]](http://i.ytimg.com/vi/Oswujxm2Ag0/mqdefault.jpg)
Timestamps (Powered by Merlin AI)
00:03-Generative AI is more significant for the industrial physical world
02:16-AI enables expertise to become more important
06:39-Higher industrialized economies are more optimistic about AI adoption
09:04 - Domain-specific AI agents are crucial for industrial applications
13:14-Capturing and encoding expert knowledge in the knowledge age
15:23 - Domain-specific AI agents are revolutionizing the industrial world.
19:32 - Domain-specific AI agents are goal-oriented with planning and reasoning capabilities.
21:22-Industrials are the early adopters of AI technology
25:30-Recurrent Loop in planning and reasoning is key for future AI agents
27:20-Incorporating recurrence and problem-solving in AI models.
31:14 - Domain-Specific AI Agents enhance problem-solving abilities in industrial settings
I really enjoyed this, especially at the end when the SOP question came up!
Lack of AI people with domain knowledge is a big problem. A lot of them know a lot about AI but nothing about banking, manufacturing, health care, or any part of the real world.
this is very true. AI skills need to be taught to the domain-specific engineers e.g. I have a much better grasp working in healthcare having done biomedical engineering, but I cannot guarantee the same translation to another domain.
@@hifzul123 Within your domain you can be recognized as an SME. The key issue in using your own AI's is loading domain-specific information to the knowledge base. Consider learning to use GPT4ALL. It lets you run AI's on your own PC. You probably want i5 or better because they run faster with more threads. I'm running an i7 with 16 threads. The Llama 3.2 1B model is a good choice for running with your own knowledge.
@@tangobayus Nope. The AI "people" don't need to "have" the knowledge. They scraped the internet and all public libraries and public papers, etc... there is no problem here. Also, it only takes one good source of data. Also, the next gen AI models will create their own data, by interacting with the world. No problems here folks.
And do what with all this AI if many of us are employed? Who can afford the good and services produced by AI?
Nice. AI surely can replace CEO/CTO/CFO positions.
Capturing knowledge is challenging. All AI agents should have a CV profile that includes qualifications, work history, experience, and feedback ratings from humans. Then we can interview and onboard the agents using evidence and reference letters.
This is exactly what's going to happen.
Feedback ratings from humans have never been reliable. Especially if it’s about something like AI? No chance.
A performance mechanism built into blockchain will track its performances per accomplishment (Agentic AI)
Fuck humans, humans are bitter and strange, said just as bitterly 😅
Well ChatGPT first told me that 2024 was not a leap year, then gave me the calculation formula and showed it was. Should we also do drug tests?
Why did you edit out the most important part of the presentation?
What was the most important past that was edited out?
Please explain.
I'm going to guess that the answer is actually within your own question. It obviously was the most important part of the presentation and too good to put online here lol.
16:54 that’s the part being edited and cut
@@tracytsaiwhat was Nguyen talking about in that cutted part? Could you please tell us :)
Ai agents will do many things for us in the future.
He said that the person that was helping them train the domain-specific expert level AI was about to retire. Exactly, because anyone who trains domain-specific AI is about to retire their human job, whether they are ready or not.
Until these companies have an answer to how they are going to honor the loyalty of these employees with loyalty FROM THE COMPANY that helps them get realistic LONG TERM, sustained training to keep them paid and relevant, the U.S. will not only remain pessimistic about AI, but expect that pessimism to ROCKET upwards as this tech is (pun intended) employed, while humans are no longer, more and more.
You raise an important point about company loyalty during AI transitions. What if companies invested in reskilling programs that help employees become AI operators and domain experts in their field? This could create a win-win where experienced workers guide AI systems while developing new valuable skills
@Custom.AIAgentAcademy Why would they do this? Did they retrain coal miners? Wake up. We are in a TERRIBLE position economically and politically for the integration of AI. Labor union power is at an all time low, mega corporate and oligarch power is at an all time high. The PTBs have been sucking every penny from the bottom to the top; AI and robots are SPECIFICALLY designed to do just that. Without a MAJOR revolution starting almost immediately, we are all F'd.
Does anyone have acess to the presentation?
Scaling innovation one algorithm at a time 🔥
Would there be any jobs for entry level fresh grads in this domain and how to get those jobs and what companies are hiring?
Would have liked to have heard more about domain-specific AI.. Poe, Causaly (science), Consensus (academic research), and Hebbia (financial services).. and chatboxes.. Character, Claude, Grok, Meta AI, Poe, DeepSeek (China)
You mention some interesting domain-specific AI tools. Have you had experience working with any of them? I'd be curious to hear how they compare in terms of real-world applicability versus general-purpose AI
I hate to be negative but manufacturers have been "capturing" domain specific process data for about 50 years. It is often written down but it is always captured and taught to younger employees. In die casting plants we have 60 process inputs and it is quite complex. I am sure that a chip fab is quite a bit more complex. Capturing the data is not new at all. Tha challenge is to know how to use the data to solve pocess problems, test the results and improve the process. This requires skilled people. AI will not be doing this any time soon.
You are correct and it’s not negative. In particular, in the semiconductor industry, we have long had extensive documentation and processes to capture operational expertise. The exciting opportunity today is that it has become much easier to capture and to operationalize using AI. Because today’s AI can much more readily “speak/understand” natural language and other sources of knowledge from our physical world.
I agree with the first part you said, however LLM does have reason capability but “training” it to consistently generating the right answer is a challenge, at the moment. For domain specific topic, it’s even harder. the training data set is proprietary, low volume, and the LLM was largely trained on noisy data which causing the result to be inconsistent and unreliable. But this can be engineered and corrected over time. It’s no longer impossible.
I disagree! It's not about AI solving the problem. AI as a tool can enable skills development using data and processes in the context of manufacturing.
@@edwardjones856 You have zero reason to say AI won't be doing it soon. How complex is programming? Making entire videos in seconds with a prompt? Imitating voices near perfectly. The AI merely needs to be trained on the correct data sets, the exact same ones that the humans would be using, and given the correct supervision and guidance during training. The presenter just spelled it out in THIS VIDEO. I'm sorry sir, but your insurmountable obstacles are EASILY solved, even at current AI levels and when AGI arrives, AGI will do ALL of the solving: period.
I'm working on a screenwriting project where we are using AI's for feedback. They can improve the writing sometimes but they can't come up with the creative ideas that we provide.
I found this a useful talk. for a non tech person looking for high level understanding. thanks.
Great to hear!
Please help!! with which ai open source models I can reach everything that was discussed in this video? 🙏
We need AI to increase the leverage of physical workers at the coal face - we need 10x the resources to get off planet and white collar work is obsolete, but smart guys with wrench’s own the future
Esoteric ..
Is this not why PLTR ontology is so important
I think I am not understanding what the speaker wants to say. Even langchain or cot or tot or agent using LLM to create plan
Yes, Is it is true that as of 2024 we cannot 100% depend on LLM. Therefore we should go slow in automation. But the reason he is giving is incorrect.
and "the stuff which is not in those documents" is so much...
Exactly. And humans keep generating this stuff.
Initially, AI is an enabler. If it emerges in an ethically corrupted milieu it will enable corruption. Alternatively, it could facilitate the emergence of a society that measures performance against commitment, personal agency, responsibility, and equitability, a “great equalizer” promoting objectively wise outcomes. This is an existentially significant choice. Read Amaranthine: How to Create a Regenerative Civilization Using Artificial Intelligence
Many of the models come from a corrupted milieu but they can be put to virtuous purposes.
US must detach herself from all types of wars, focus on innovation and it's mass scale use to create competitive edge.
US wars from the past 30 years have only benefited one nation and it hasn't been the US. Unlikely that the US will end any wars anytime soon since every party is captured except the green.
remind me where would you buy microchips without Taiwan?
@@Dmytro-kt3frexactly why America needs to start manufacturing again
@@The_Quaalude lol, good luck. All of that tech is proprietary to Taiwan and Taiwan only. Taking that your disabled country has constant leaks of military blueprints to chinese there is doubt that States are able to do anything at all. Nobody is scared or even respects usa rn.
And idiots that scream about isolationism just continue to prove that nobody should believe in this failed state
The biggest risk the US has is that almost all the Semiconductor company have their design centres in Israel as well. Now, any electronics have a very high risk of getting compromised, including turning into a bomb as we saw in Lebanon.
I just don’t see how the pace of new knowledge creation does not end once the humans are all laid off or retired. Endless permutation of old by AI is not creation of the new.
Spot on. They are great for regurgitation but the sparks of creativity come from somewhere else. God or whatever you wish to call it. You need a soul to get information from God. AI's don't have one.
Hi please come see me how we are using agents to build 200,000 word legal contracts
Send the link
You are on the right track. I built a GPTs with 41 scientific papers about the Covid spike protein.
I wonder if in the 'expert mode' contexts of GenAI, when we'll stop calling the over-confident, incorrect (and even made-up) responses 'hallucinations' and finally begin to call them 'bullshit' as we would call out GenAI's human analogues. But wait, GenAI has 'PhD level nunchuk skills'. We're talking super-PhD-impossible-to-bullshit level mega brain (with weaponry! ) here. Nah, why question 'it' either - that's far above our mortal pay grades ...
yet he provides no evidence that they work better. just claims. how is this remotely scientific.
I think you confused about what this is.
only the domain expert in engineering are able to implement the AI . IT and CS people have limitation
the only ones who will benefit from AI are the rich
The educated
Absolutely wrong. You can learn a lot from them, and that makes you smarter.
I found this presentation strange. I worked in the 90s creating expert systems. It failed miserably. Why?
How we do things today is already inefficient and out of date. Adoption, standards, expertise are all behind R&D and emerging know-how and innovations.
Experts are actually the worst at predicting the future and at adapting to the future.
Every study done on this topic comes to the same conclusion: experts can't see their own biases. They are slow to change their minds in a rapidly changing world.
Agentic AI is important. It is the top layer of AI, but it will grow organically from data and self-reflexion over time.
Today's human specialists are tomorrow's dinosaurs.
I too built some expert systems back then. They suffer from combinatorial explosion. I think they have come back in the guise of robotic process automation. Generative AI is not good with executing processes reliably.
Interesting perspective from your 90s expert systems experience. How do you think today's machine learning approaches differ from those earlier expert systems? While experts may have biases, couldn't their practical knowledge still be valuable in training AI to handle real-world edge cases?
US is No. 1 in innovation but China Master it's use it then build upgraded version.