1 00:00:00,000 --> 00:00:01,000 Hello guys. 2 00:00:01,000 --> 00:00:06,000 So in this video I'm going to talk about the evolution of large language model. 3 00:00:07,000 --> 00:00:13,000 Right now you may be hearing about various LM models that have been created by different different companies. 4 00:00:13,000 --> 00:00:20,000 But to start with, you know, large language models probably initially came up in 1967 called as Eliza. 5 00:00:20,000 --> 00:00:26,000 And this was specifically used to solve some of the LM, uh, some of the NLP use cases. 6 00:00:26,000 --> 00:00:32,000 Then in 1970, we had this another model, then Then in 1918, we had this Excalibur. 7 00:00:33,000 --> 00:00:42,000 And in 1988, from here, the advancement started happening, okay, which is with respect to the simple 8 00:00:42,000 --> 00:00:42,000 RNN. 9 00:00:43,000 --> 00:00:46,000 Then in 1997 you had this LSTM. 10 00:00:46,000 --> 00:00:52,000 And if you don't know, in 2014 we had the GRU RNN, right. 11 00:00:52,000 --> 00:00:58,000 So in the LSTM RNN here you could see that we had a feature that was added with respect to long term 12 00:00:58,000 --> 00:01:01,000 memory and short term memory. 13 00:01:01,000 --> 00:01:09,000 And later on in 2017, this was the amazing breakthrough wherein an attention mechanism was basically 14 00:01:09,000 --> 00:01:13,000 added in this architecture, which is called as transformers. 15 00:01:13,000 --> 00:01:20,000 So after coming into transformer in the next year itself, we got Bert and from this we got this amazing 16 00:01:20,000 --> 00:01:21,000 GPT models. 17 00:01:21,000 --> 00:01:28,000 Okay, now, uh, as we go ahead in 2019, you got GPT two, Robert. 18 00:01:28,000 --> 00:01:32,000 Uh, and in 2020, you found out GPT three, right? 19 00:01:32,000 --> 00:01:38,000 And that is where your open AI right now many people are specifically using ChatGPT. 20 00:01:38,000 --> 00:01:38,000 And. 21 00:01:38,000 --> 00:01:39,000 All right. 22 00:01:39,000 --> 00:01:45,000 So OpenAI probably came up with this amazing chat GPT applications by using this GPT three. 23 00:01:45,000 --> 00:01:49,000 Then in 2021, they moved to GPT 3.5. 24 00:01:49,000 --> 00:01:51,000 In 2020, are still. 25 00:01:51,000 --> 00:01:55,000 Google also came up with palm models Instructgpt and ChatGPT. 26 00:01:55,000 --> 00:01:58,000 So these are some fine tuned models itself. 27 00:01:58,000 --> 00:02:03,000 And if I probably consider with respect to 2023, meta also came into picture wherein they brought their 28 00:02:03,000 --> 00:02:05,000 version of llama models. 29 00:02:05,000 --> 00:02:11,000 Then uh, again, OpenAI came up with this GPT four, then palm two was their Bard, was their Dall-E 30 00:02:11,000 --> 00:02:15,000 was there, and many other models were specifically there, like Falcon. 31 00:02:15,000 --> 00:02:21,000 Um, going forward, you know, right now the advancement are still happening because if I probably 32 00:02:21,000 --> 00:02:22,000 consider 2024. 33 00:02:22,000 --> 00:02:29,000 So in 2024, OpenAI has already come up with this amazing model, and this time it has come up with 34 00:02:29,000 --> 00:02:30,000 this multi model. 35 00:02:30,000 --> 00:02:35,000 And the multi model is nothing, but it is GPT four. 36 00:02:35,000 --> 00:02:36,000 Oh okay. 37 00:02:36,000 --> 00:02:41,000 And with this you will be able to perform tasks with respect to both NLP and images. 38 00:02:42,000 --> 00:02:42,000 Okay. 39 00:02:42,000 --> 00:02:45,000 You will be able to even generate the images itself, right? 40 00:02:46,000 --> 00:02:49,000 Um, along with this, uh, Google will not be left behind, right? 41 00:02:49,000 --> 00:02:55,000 So in 2024, Google already they have come up with its Google Gemini Pro right. 42 00:02:55,000 --> 00:02:56,000 Google Gemini Pro model. 43 00:02:56,000 --> 00:03:00,000 So Gemini Pro 1.5 right. 44 00:03:00,000 --> 00:03:05,000 And along with that, Google Gemini Pro 1.5 flash is also there. 45 00:03:05,000 --> 00:03:07,000 Uh, that specific model is there. 46 00:03:07,000 --> 00:03:12,000 And over here also we have some round three variants which you can go ahead and serve with respect to 47 00:03:12,000 --> 00:03:14,000 the Google Gemini models itself, right. 48 00:03:14,000 --> 00:03:20,000 Not only this, uh, in 2024, uh, there are many, many companies who are specifically coming up with 49 00:03:20,000 --> 00:03:21,000 this, right? 50 00:03:21,000 --> 00:03:22,000 Uh, llama three is also there. 51 00:03:23,000 --> 00:03:25,000 Then, uh, you saw about anthropic, right? 52 00:03:25,000 --> 00:03:29,000 With respect to anthropic, you have this cloud cloudy opus, right? 53 00:03:29,000 --> 00:03:30,000 Cloudy opus models. 54 00:03:30,000 --> 00:03:31,000 Right. 55 00:03:31,000 --> 00:03:37,000 And um, uh, along with cloudy, you also have all these models in hugging face also. 56 00:03:37,000 --> 00:03:40,000 So everywhere this is specifically present. 57 00:03:40,000 --> 00:03:43,000 Even in hugging face you'll be able to access this particular model. 58 00:03:43,000 --> 00:03:48,000 And right now, recently I guess uh, Google Gamma two model has also been coming. 59 00:03:48,000 --> 00:03:50,000 And this is an open source model. 60 00:03:50,000 --> 00:03:50,000 Right. 61 00:03:50,000 --> 00:03:52,000 So Google Gamma two. 62 00:03:52,000 --> 00:03:57,000 Um, if you if I probably consider Google Gemini Pro, this is a paid model itself. 63 00:03:57,000 --> 00:04:03,000 And uh, for some for some hits or some, uh, request, this is given as a free. 64 00:04:03,000 --> 00:04:08,000 But if I consider, uh, opening AI, GPT four or these all are paid models specifically, this is an 65 00:04:08,000 --> 00:04:09,000 open source model. 66 00:04:10,000 --> 00:04:16,000 Uh, so this is just to give you an idea over here, like what all models were there and the evolution 67 00:04:16,000 --> 00:04:22,000 of this large language models, uh, it's like coming from 1967 with a lot of research, a lot of development 68 00:04:22,000 --> 00:04:24,000 that is specifically happening now. 69 00:04:24,000 --> 00:04:29,000 What we are going to basically do is that in our next video, we are just going to compare the speed 70 00:04:29,000 --> 00:04:31,000 and accuracy of all this kind of models. 71 00:04:31,000 --> 00:04:36,000 And, uh, it is very much important to probably know, like which model is best and how the model are 72 00:04:36,000 --> 00:04:40,000 specifically doing, uh, in the industry, which people are specifically using it. 73 00:04:40,000 --> 00:04:41,000 Right. 74 00:04:41,000 --> 00:04:44,000 And that is what I will be discussing in the next video. 75 00:04:44,000 --> 00:04:45,000 Thank you.