Introduction to modern AI

0:00 And if you are serious about building a strong career in AI, check out [music] Scala's advanced AI and machine learning program. It's designed to help you master the skill needed for the real world AI role and get you ready for that. For more information, check the link in the description box. Also, tell us in the comment what other AI topics you would like us to cover. So, let's get started. So, this is the first video of this series and um this is the only video where we are going to talk about theoretical concepts.

0:45 Now, uh this particular series is going to be really practical. uh you will be building a lot of applications together with me. But before we can jump on doing all those amazing uh things and building up all those amazing uh applications, it is very much important to understand the evolution of the entire AI space. So how chatbot became chat GPT and how it transformed itself to a gentic AI is what we are going to talk today.

1:10 So let's get started now. First of all let's understand how a chatbot functions. So we will talk about chatbot and guys believe me it is going to be really fun to understand the entire evolution because when you understand the evolution you understand and you respect the powers that we use today so it is going to be really fun just focus so when we started with and I I'm taking you to the time which is around 2018 around 2018 there was a great buzz about this term which is chatbot and a lot of people were fascinated about how chatbot functions.

1:56 So at this particular time you will see a lot of attraction coming towards these chat bots and chat bots were becoming really powerful. Uh all the businesses, all the tech giants were trying to implement chatbot in their applications and uh you would be seeing at that time all the websites were also enabling their website with this chatbot to help customer interact uh with their website and get their FAQ answered. So a very simple example of this chatbot can be seen over a website or let's take a example of a banking website.

3:11 So let's take HDFC bank. HDFC bank. Now when you go to this website you will find out this small popup coming from the right side which is basically known as the chatbot. So this is your chatbot. Now if you ask this chatbot a very simple question and that is known as a query. So remember we are going to use this term throughout the series. A query is a question from the user. So that is what a query is.

3:36 So let's say the query is how to open a bank account. Now this is a very simple general question right now how this chatbot will resolve this query. So this chatbot will take this question or the query and the bank would have trained this chatbot on a series of questions and answers which are kind of the repository of FAQs. So this is FAQ repository where they have preserved all the questions and their suitable answer. Question suitable answer.

4:25 Question suitable answer. Question suitable answer and so on. Question suitable answer. Now what this chatbot will do? This chatbot will take this query and will go in this particular FAQ repository to find out which particular question in the repository matches to this particular query. So what this chatbot will perform is known as the similarity matching. And this is a very simple idea behind finding out a similar question related to this query. So it is simply finding out which of these questions are very much close or similar to this query and that is what this chatbot was doing.

5:14 Now let's say it was able to find out that this question is very much similar to the user's query and that is what this chatbot will return. So this chatbot will take this answer and will return to the user this answer and that is how the traditional chat bots were functioning. They were not doing anything amazing. They were actually looking at a database and they were finding out the similar question using similarity matching and were returning you the best match with the corresponding answer of the best match.

6:03 Now in this entire process if you ask another query let's say another query is who is the PM of India. Now this is the question. Now what do you think that is this chatbot capable of answering this question? Is this chatbot capable of answering this question? The chatbot will return you the generic answer because this chatbot will not be able to find out any answer related to this question. Any question in the FAQs or the database which is similar to this query.

6:55 So your chatbot or this chatbot will return a generic answer and the generic answer would be um unable to understand your query. Should I connect you with the um customer representative? So this is what the generic answer that you will get out of this particular chatbot. Now you guys would have understood one thing that these chat bots were very simple machines and they were designed with two constraints. They were designed with two constraints. Number one constraint is they were doing what is known as the similarity matching.

8:12 They were not generating a new data. They were not trying to give you something which is very new. They were simply doing what is known as a similarity matching and that is what we have seen. The second thing which they were limited with is their domain knowledge. So these chat bots were designed with a limited knowledge to function in a limited domain. So to function in a limited domain and what I mean by domain I basically mean that this chatbot can answer your questions related to a company in which it has been trained on.

8:59 This chatbot can answer questions related to a specific industry also. But the point is that this is not a generic chatbot. This chatbot can only answer your questions related to a specific industry or a specific company. They are not generalized chatbot. And that is where scientists started thinking can they somehow convert this domain specific chatbot to a generic chatbot or a generalized chatbot. So that is where started the next phase of innovation. So I'm just marking this.

9:50 So this was the start of the next phase of innovation. So scientists were thinking about it. The objective was that they want to come up with a generic chatbot that is not limited to any company or domain. So they wanted to come up with a chatbot that is generic. Generic means a chatbot which can answer questions related to any industry. A chatbot which can answer questions in any language. A chatbot which can answer questions related to any specific genre.

10:41 So that that is what the scientists were thinking about how they can come up with a generalized chatbot. But there was a problem. If you want to come up with this generalized chatbot and you have to train this machine on a huge set of data. So this is huge text data and that is what you need to feed to this machine. So huge text data means the open internet data opensource internet data and that was the problem which the scientists were facing.

11:29 So their objective was clear. They wanted to come up with a generalized chatbot which is not limited to a company or a domain. They wanted to come up with a a chatbot which can answer your any general question. That means they were required to train a machine on a huge set of textual data or in simple language they were required to train a machine on the opensource internet data. Now the problem with this kind of mechanism was that there was no such algorithm or the technique to make this machine learn from this humongous amount of data.

12:17 And that is where this scientist whose name was Ashish Vaswani. So in in 2017 I think uh Mr. Aishwaswani wrote a paper and the paper was named as attention all you need. Now Ashish Waswani in his paper introduced two concept. one was the attention mechanism and one was the usage usage of transformers. Now those of you who are thinking what is this? What is attention? What is transformer? In a simple plain English these two things are the techniques which helped the scientist to carry out this process.

13:08 This was the process which were the which was getting uh which was not getting done because there were no such technique to train a machine on such a humongous amount of data. So since the machines were um since the scientists were thinking about training this machine on a humongous amount of data but there was no such machine learning technique available. Aishwasani in 2017 wrote this paper named attention all you need and that paper introduced this concept and this concept which created the base of this entire process.

14:00 So due to him due to his paper scientists started thinking that how to enable machine learn from humongous amount of data and that created the next wave of innovation in the industry. So using his paper and his idea, Google and one more company open AAI these two company started working on his paper and he started implementing his paper. They started implementing attention mechanism. Google were uh working on their project name Google Bard B A R D Google Bard while OpenAI was working on GPT.

14:48 So they launched GPT1, they launched GPT2. But these two launch which is GPT1 and GPT2 were not the public launch. So they were experimental and they were not launched to public. Not launched to public. But in 2022 and the date I remember is 20th November or 22 November. So 20th November 2022, chat GPT was launched by OpenAI which was basically GPT3 and this was the very first time the entire humankind was able to get a UI a user interface where so this website was called as chatgpt.com And here you were able to get a interface where you can interact with this generalized chatbot.

15:57 So this was basically the generalized chatbot. So this particular GPT3 which they came up with was nothing but the LLM that we know today. large language model and this is what the machine I was talking about. So these machines are basically the LLMs large language models and why they are known as large language because language because they were trained on the textual data the language data large because they were trained on the humongous amount of data and these are machine learning models that is the reason large language model.

16:49 So they were large language model and let me tell you one more thing Ashish Waswani at that time was working with Google minds as a research scientist and that is when he wrote this paper. So I hope you guys were able to understand the entire evolution of how chatbot became chat GPT and there was only one requirement how to convert this domain specific chat bot to a generalized chatbot and that is how