Take a Byte | 科技咬一口

EP3 | 一起進入 AI 的魔法世界 Feat. Vito

Take a Byte Season 1 Episode 3

AI 從去年就開始大爆發,每過一陣子就會有很多新的 AI 工具推出 🤖 如果對 AI 工具有興趣的話記得去追蹤我們的 IG 喔!大家都在談 AI ,但是對於 AI 好像有點了解卻又說不出來到底是甚麼,這集節目就和我們一起進入 AI 的魔法世界中 🪄

本集主題
• AI 以及背後的原理
• AI 真的是吃電怪獸嗎
• 人類會被 AI 取代嗎
• LLM, LAM 是甚麼? 區別又在哪邊
• AI 和自動化、智能化是同一件事嗎
• Machine Learning 是怎麼訓練的

🚀 記得訂閱我們的節目,每個月都有新的內容等著你發現。Take a Byte,讓我們開始這場數位之旅吧!🌟
Follow along with English subtitles on Youtube.

與我們聯絡~!

Support the show

感謝您的收聽!想要獲得更多更新和內容,請關注我們的社交媒體頻道:
☕ 訂閱 TaB+ 加入我們的專屬Line群 | https://www.buzzsprout.com/2307755/support
🔗 關注我們的社交媒體頻道 | https://takeabyte.buzzsprout.com

Hello everyone, welcome to Take a Byte In this program we discuss IT in a casual chat Netcom technology and industry issues You can also watch our subtitled content on Youtube The link will be in the description below So what we are going to discuss in this episode is AI, because the content of AI will It’s a bit more, so we should divide it into two episodes like this AI has exploded since last year, but Every once in a while, a new AI tool will be launched Then if you are interested in AI tools, you can remember Put our IG in it Then everyone is talking about AI, which may be applied in daily work. I seem to have a little understanding of AI but I don’t know what it is. So in this show we have Vito Be our guest Welcome to our shelter Um, no, it's our show Okay, then we can ask Vito to introduce himself to us first. Hello everyone, my name is Vito and I am engaged in computer information. The coursework has lasted for more than twenty-five years. Well today, AI is quite powerful. This has been very intensive recently, especially from 2022 onwards. So today let’s discuss with you what is AI? Thank you Vito! It means Vito, right? Some of our colleagues said If you are a French person, you would pronounce it Vitton. The name comes from Italian I actually did a brief search to find out what life means. I didn't look it up carefully. There is a very old movie called God Father. It turns out that the main character above is called the gang boss pair. He is very domineering. I think this name is very domineering. What are you talking about about him? Anyway, at the beginning of the chat show about movies, we can talk about a few words. Then sit on my left He is also a good friend of ours named Johnny. Can you introduce yourself? Hi everyone, I'm Johnny I am honored to participate in this casual chat on this program today. That's because Vito's side Very specialized and professional experience in the field of AI The main thing is to chat with everyone and talk about what AI is. Is it magic? Ray just mentioned that 2022 is very popular. I want to say whether it is because of the end of 2022. At that time, I felt that Open AI’s ChatGPT This thing they launched was very popular at first, right, and then many people wanted to say, "What is this?" We can pass a sentence to him and he will help you translate it and write it into an article, like this In fact, I had already used AI before going to ChatGPT. Then use AI to generate pictures and then at that time It is very popular to convert each picture into another style. OK Already waiting before ChatGPT Exactly When sorting out each paragraph, we used ChatGPT to convert it into Mandarin. Because it’s a completely new concept, it’s awesome. It’s no longer like talking to Siri Siri seems a bit idiotic now Will this be blacklisted by Apple? They should also feel it in their own hearts, so I believe they should also be developing their own That's right, module by module. Well, why not come with us? Simply ask Vito to explain what the concept of AI is. How to explain AI? So what other situations are often misunderstood by everyone? Okay actually AI For those who are not familiar with it, AI is actually a very abstract concept. What he has to do is to make a so-called computer what we Taiwanese call a computer Then he will imitate some intelligent behaviors of humans. For example, we used to see ads Computers are just for picking potatoes. We are not qualified for this An AI vision is an effect on visual recognition Well, for example, Alphago should have been released in 2017. If a Go program beats the world's top players, it is also an application of AI. The field of AI is actually quite broad at present. AI is relatively common There are many misunderstandings in cognition. I think they can be divided into three aspects. The first one is that AI will replace your job Well, I think many people will have this worry because some people will If you confuse AI with automation, then for now What he can actually do is quite limited and In fact, I want to propose a more innovative concept That is, you don’t actually need to worry about your job being replaced. The problem is that every generation since the industrial revolution to the present In fact, new and different professions are constantly being created in every generation. Just like sixty years ago, there may not be roles and positions for hardware engineers like us. that was created after the information revolution and that the opposite AI will also create new jobs and eliminate some old ones. But the world is rotating, and the world keeps turning In fact, we don’t need to worry about our work disappearing, so it can’t be imagined as saying Artificial intelligence is an auxiliary tool that helps us. It can be called imagination. Yes it helps you be more efficient and there is some information that will help You can make more accurate predictions and help you make future forecasts, etc. Because things like Microsoft’s Co-pilot are not yet popular, but I have to wait until now. After that, it is for the enterprise version, but if it can help, I think I can help, for example You enter keywords and it can help you find some information you wrote in Word. It’s quite special to sort it out directly for you. I remember that it can be reorganized and then watched again Its goal is to help everyone reorganize it and then directly start from Convert the text into a PPT presentation That's one of its functions In fact, it should be good news to us ordinary office workers, because sometimes he can't tell you Please help me find what reports I need for 2017 and 2022. It’s a pain to sort out the reports. But if we make good use of AI, will it mean that we can use comparable and more efficient methods? I think efficiency is the key point. 我觉得第一个比较显著的改善就是在效率啊 Well, the second common misunderstanding is that AI has super intelligence. It will destroy humanity just like in the movie some thoughts of fear In fact, from my point of view, AI is still very far away from reaching that stage. The most practical question is whether we have enough energy now to cope with it. The fact that AI is booming to that point is one that I personally think is the biggest problem. Many people say that AI servers eat power monsters, but this term Yes, actually with the current AI technology I personally think it probably can't be done Most of the things in our daily life include autonomous driving. It can really simply assist you on the highway. Other self-driving functions are still far inferior to those of us humans today. Driving effectively is especially so. Taiwan is traffic and our roads are special. What's special is that some people will suddenly rush out. Or riding and driving indiscriminately Is it the mechanical public enemy? So old Arnold Schwarzenegger a lot of what RoboCop? (Terminator) It's not like it's a mechanical policeman I really don’t want to admit that I’ve seen it Anyway, it’s the Robot Police who are back. They are all movies about robots destroying humans. Many people used to create a myth and thought it was right. I think the more famous one is the one I will be back. to Arnold Arnold's right Because the world in the future is ruled by robots and humans, yes, yes, I hope they can coexist. We are now all required to work together with other ecosystems Let’s get along with the robot later I think that one is still too far away from us. There should be more We can't do it because we are curious, that is, here I want to ask a question to Brother Vito. What we are talking about now is like what Ray said just now, what AI will you use to draw? Then some of the things we just mentioned are so-called generative AI like ChatGPT. I remember saying that now I remember that the term for the next stage after generative AI seems to have become "let him" Develop one that is as close to human intelligence as possible What are the names of that thinking and system? Strictly speaking, I don’t know the name myself. Not only do you know what he used Yes, it is the language architecture of LAM Yes, because what you are talking about is the term that has not been concreted yet because originally It is a big language LLM and now it becomes LAM A is Action I think it will take some time for LAM to become a standardized architecture Well, for example, there are some new things, such as the AI ​​Pin that we talked about recently. It's the one who uses his palm to control that thing and says it's about to be shipped. We don’t have any karma to collect it, nor do we. Then we are just curious about this thing and this behavior. Because you use movements to control it, is this considered LAM or LLM? LLM can do it In fact, LLM can do a lot of things Then we Speaking of LLM LAM, you can also understand a little bit Let’s take a look at how it helps with learning. Features of Intelligent Learning In fact, both of the functions of intelligent learning belong to A programming language framework that its The function is to carry out a large amount of data in large quantities. Process and analyze to generate new data that is close to the original data A new information comes out He can then generate new data based on the new data and slowly as close as you could get to the goal you set out to achieve at that time For example, let me tell you a simple wordingFor example, I need a very straight road with two buildings. It happens to be when the sun is setting next to me Maybe he can generate it based on these wordings When a picture comes out, what he did is based on what I said. He went to find a lot of information to analyze what the building should look like. His building looks like the building you imagined so this is LLM, right? What does he mean? What's his abbreviation? We should tell the audience about it Large Language Model Large Language Model That's right, that's right. LAM is a large one. Action? Chinese translation is action Large Action Model Yes, then it becomes say after your next instruction LLM is when you give an order and then it may produce What you expected The output result may be text, just like a picture or an article It’s LLM. For example, you just mentioned it at the beginning I may need to define a topic so that he can write a paper Written as a paper, this is a target For example, what I need is to turn it into a picture Becoming a picture is my ultimate goal It will slowly generate the information he wants to search based on your goals. To generate all the elements he needs for this picture Generate it as you said What's the difference between his final results and LAM? LAM actually doesn’t have a so-called killer yet. Most of the applications are still in the concept stage. Well as I said it probably won't be for a while yet For mainstream, it must be similar to ChatGPT When a relatively large practical application appears I think we will have a better idea of ​​what it will be like in the future So if we use imagination, we can imagine what LAM can do for us? For example, what kind of actions does it help with? In fact, LLM can do this. Because in fact, like a robot, if you give instructions, it will follow your instructions. Do it or something like that It can perform these actions according to your instructions. But LAM I think it would be a little richer Don't say you might sit down it might be in various Different ways to do it more detailed But yes You tell him that there was an AI Pin before The orange one is right rabbit red for a while I think the biggest difference is what you just said For example, if you want to book an Uber, you can After you give the command and know that it can be automated, he can execute it by himself and then book Uber. If the subsequent action is that of LLM, he can only It seems to help you open this application Or does it mean there is no execution end behind it? I should say then LLM can help you do the execution of the Booking Uber takes action, but LAM will do it more thoughtfully When do you need Booking and where do you want to go from? It will derive a lot of relevant information you need When you come out, he will do more precise movements to help you book. In fact, after hearing this, I think everyone should have a concept, which is like saying that AI is really making people Our life becomes more convenient and we don't have to worry about it. Can't we call others in the future? Will AI replace me? Will it be like WALL-E? Just like WALL-E, everyone has become useless. Oh my God, you're showing your age. I'm not worried about showing my age I'd like to ask if it's Vito or Johnny? Do you usually use AI to help you perform at work? In fact, I think the most commonly used It’s ChatGPT, because sometimes I need to read some because of work. Some technical text and I don’t know much about it In fact, I just throw the text to it and I think the quality of its translation Maybe compared to Google, I think ChatGPT The quality of translated books is closer to what we people say Because compared with Google, ChatGPT feels that it is translated by a machine, but ChatGPT It's more like how people treat it. If you encounter some words you don't know, you can ask it because of course If it is not paid, it is now version 3.5, and the 3.5 version is in 2021? So of course it's relatively new and doesn't know about 3.5 and 4 I really don’t know what’s the difference. I’ve never used it. I don’t even have a ChatGPT account. I think Brother Vito doesn’t seem to be here. Few AI tools are used, mostly for research and development. Just because R&D work requires more communication between people. Communication between people rarely requires the use of AI So in fact, we can say that AI has some things that it may not be able to achieve, like us This kind of warmth between people is because we have emotions and expressions There is no way for AI to be so merciful to people right now because it is still not up to this point. Yes, that's right So Brother Vito is doing AI but he doesn’t use AI. Too much contact and I'm a little tired of it Just like a chef will not always eat food from his own restaurant I won’t say that I want to cook something when I go home and order takeout myself. I should say that actually every company Has been launching its own new AI Solution I always try the trial version first. Try it to see what it can do Generally understand what it can do because actually It has quite a lot of restrictions like Google Google launched Gemini last month Because the one at Google was originally a Bot and then Gemini replaced him. But once it was launched, I found that it had too many problems, such as what? What will you find? Many of its understanding behaviors are quite stupid. It probably doesn't have enough information It’s not done well enough and it should be used by multiple people. After just offending Apple, now it’s offending Google. Oops, he's going to be taken off the shelves now. Gemini is off the shelves now? It will be used up by then. It was taken off the shelves in less than a week because it seemed like it would say a lot. I remember there were some forums Posting it, it said some very strange words, some words that would harm morals. In fact, Microsoft used OpenAI a lot from the beginning. Open AI suffered this loss before in 2006 As soon as I posted it, he fed me some information about racial discrimination. It means that many people can test his bottom line. I think the same is true for Gemini. There are many people who like to play these memes and then they Google Maybe I haven’t learned it yet, so I actually want to listen to what Brother Vito shared. I would say that the training process of AI is basically that children grow more and more from the age of 0 to 6 years old. I used to only write one word in the book, but then we had some words and I just spelled it.

An article:

Is the learning process of AI like that of a child? We think he is still a little baby, right? Is he a little baby with very strong learning ability like this? In fact, his growth process depends on how much information you feed him. Well, just like that we just mentioned AlphaGo He threw some chess records to him The result will definitely be different from throwing a thousand chess pieces to him. What is Alphago? Alphago is There is a company called DeepMind that can play Go Defeated, defeated the world's number one chess player. Not only did he defeat He played against a Korean and he won 600 games I have never lost any game against players above 8dan. Okay, let's change the topic What I want to ask is that AI and automation are the same thing. What's the difference? I think many people have misunderstood AI. In fact, automation can be related to AI. combined but automated to me automation is automation What automation often does is help For example, in human factors engineering, we work in factories Some operators may do one thing at the same time He will have people who can't stand for long due to engineering problems. We may be able to rely on automation to help them solve this problem You don’t need any AI, but you can probably rely on AI For example, our factory has some packaging inspections that require AOI. It automatically helps you do these so-called Automation plus the function of AI, AOI is completely different from AI AOI is an optical inspection because it will tell you when the light source is detected. He only tells you that it is like this now. Before, people analyzed them, but after AI, they were you. I just want to tell you why there are some yield problems here or, for example, the silicon is not polished after the board is finished. Or the analysis may tell you the reason. This is correct to understand. In fact, just as you said before Now many people will misunderstand that Misconceptions that AI will take over our jobs Yes, it is a bit related. In fact, AI and automation also have The different thing is that I should say different fields, but the two can be combined. Well, because we all have been talking about AI, what else is there? One is called ML Machine Learning Well, it’s machine learning. So what kind of concept is it? Are they different? Machine learning is one of the fields of AI, but now the current AI Most of the fields are developed in the field of machine learning. The earliest you can go back in machine learning By 2006 to 2007, when this concept was proposed Because let me give the simplest example to explain what machine learning is We just mentioned Alphago, it can play Go In the past, we might have had to have someone manually input the chess records. Tell him to teach it but this Spending labor time is not cost-effective, right? The overall efficiency is not good Machine learning is to let him read the chess records by himself Don’t tell it, don’t tell it the rules, let it learn automatically. Let it be judged What is this rule? What is winning and losing? But isn’t this supposed to have a high rate in the beginning because it can’t understand it at all? That's right, that's why I said the amount of information you have will determine your final result.

Let's take a simpler example:

We identify the numbers 123456789 as Arabic numerals Then someone wrote it very standard, just like our license plate. It's very standard, but someone wrote it in a messy way. So how do you identify it so that even a machine can identify it? What number did you write? This is actually the work of machine learning! It suddenly occurred to me that because you told him, it was the part where he wrote. Like many generative AIs, their biggest bottleneck is when generating pictures. It is to generate pictures of words and writing. If there is text in the picture, it is correct that it cannot be generated! I think the current algorithm cannot cover these parts. This requires quite a lot of algorithms. Actually I agree Ray said that sometimes, for example, I would like to ask him to generate a Japanese-style one. Street, but you can see that the writing on it is obviously not Japanese. It's a string of Martian characters on that signature road name The names on that street sign are all in Martian script I think he took the picture because he also obtained it from various databases on the Internet. Went up to grab the information he needed, but he couldn't understand it. What does the background of your picture mean? But actually I One thing I don’t quite understand is why they can already understand it. What does the etymology of the text we enter mean? That's why he can't understand this I think everyone has a misunderstanding about AI based on the words on this picture. That is, he can 100% achieve the desired effect you want. In fact, the accuracy of AI is what we have been discussing because in fact Why AI is ineffective in many industries Comprehensive import needs should be said to take a long time Keep making adjustments. This adjustment process is actually improving the accuracy. Accuracy is actually a very important factor for AI Many software companies currently invest in this area So I really want to give you a heartfelt advice, don’t think What you perceive will be the same as what he perceives Because what he calls machine learning comes out of what he calls Japanese style. Maybe he is a wooden house and then there will be He thought that some special and special landscapes were Japanese style. It’s different from how you always imagine Japanese-style temples. There are many models, but each model has different cognition. Of course I hope they can learn from each other in the future Make the entire AI database richer and more accurate We also hope that because it actually involves each family, it will be divided. I think it depends on the profitability of each company. They also want to make money, so they do To protect their own assets, they each develop their own language models. Thank you everyone for listening to today’s Take a Byte Tech Bite For those who haven’t subscribed yet, please take a second. Hit that subscribe button You can also listen to our programs on major podcast platforms Click the link below to follow our IG or search Get a byte BYTE You can find us! Well done! Looking forward to seeing you next time~ Goodbye

People on this episode