Mistral SHOCKED the industry with Le Chat! [Complete Tutorial]

Key Takeaways

hey everyone I just wanted to make a quick video about mistel I wanted to show you guys how this company has been able to produce the fastest chat experience in AI currently we'll look at that in a second but first I just wanted to tell you a little bit about mistl so basically a company from France building models for a while now some of the best open- source models these guys were one of the ones that were creating open source models from the very beginning at this point at least from what I understand is they have introduced their chat platform which is called L chat from where you can essentially use M's model just like you would with any other AI like chat jpt you would be able to chat with their models and also they have something called La platform which is their platform for creating AI agents and stuff like that we're going to walk through the lay chat first but I wanted to show you a really cool use case for using the platform here as well so and also I want to show you the pricing so for most things you can do the 0 plan you'll get pretty much everything that you would get in any other AI models for example but also you get the flash answers which are basically what makes it right now like the fastest output model available because what they use is their you know high performance servers to quickly process and give you their information so we obviously see it in action it's really nice that they have a student plan as well and you know for that you get no limits on pretty much most things and so that's that right so let's jump into lchat so in order to do that you would go to products and then lay chat and then once you're here you would want to go try lchat here right so once you click here You' be asked to sign up for an account and then log in just like any other application and then you basically land on this page which is their chat platform here you have the sidebar on the left where you can start a new chat obviously you have the upgrade button you can search for chats and you have all your chats here and this is basically where you would chat from now if you want to access the super fast feature that they have which is called flash answers you would want to go here enable it right here once you have that turned on type in whatever question you have so I'm going to go ahead and hit send so it took about 2 seconds and then pretty much came up with the whole blog article so when you use chat GPT Claud or any other AI you'll see that the output comes letter by letter so that's called a streaming response you would see it start streaming from here and then step by step output every single letter and eventually it'll get to the end and it'll take at least 20 seconds or whatever but in this case it basically just pre-processed it for a few seconds and just immediately came up with the output so here you can see if I hover over here it says flash answers this message was generated using our most powerful servers I'm just going to try it one more time I'm just going to say revise this and make it longer and hit send so as you can see right now it's streaming it so this is it being streamed just like how you would normally see AI which CH GPT or clot or any other AI that you would use so this is clearly taking a fairly long time and at this point as you can see it didn't use the flash answers so this is what I wanted to show you so the flash answers is only for the first time you chat with it obviously they're not going to use their most powerful servers for every single chat or recurring chats if you want the quick answer for something it's going to give you the answers right away but then if you want to keep chatting with it it's going to use their normal servers to come up with the output just like with any other AI now along with this that you can upload files here if you want you also have access to Canvas this is the same as what Chad GPT has you have web search you have image generation code interpreter you can have all of these turned on while you're chatting with it so to show these let me open up a new chat so here I'm going to use the canvas and I'm going to say write a block article and AI agent this is how the Canvas Works you basically have this whole dock style output where you have the whole article and then you can use this chat to just kind of incrementally improve it add things to it you could select a section and then basically ask it questions or you can ask it to analyze proof read summarize more bullet points fewer bullet points there's a bunch of things you can do directly from here or you can just like I said 's the chat as well this is very similar to what chat GPT has so I'll move on to the next one which is basically using the web search so what is some of the AI news from today that's the result it came up with you can click here to see all the sources here so it's shows you how many seconds it worked on it also showed you the steps it took just letting you know these are not used using the flash answers this is just streaming like how you would with any other AI so for the image generation I'm just using a burger in space not bad but the best part is it took 3 seconds to generate this image so that's pretty good so you know feel free to use this okay so now for the Corde interpreter now the thing about mistl is it doesn't really have a claw or a quen type of code interpreter where you'd be able to like play any code and that's what code interpreter basically means that it will generate the code and then can sort of like play itself in an environment and then just give you the output as well but what it can do is it can use Python it can use very basic things like you know generate visualization from data and stuff like that so just to show you how it works I'm just going to say generate some sample data and visualize it using python so there you have it as you can see it generated some data in the background and then plot and visualized it so if you want to see the code you can click here and then that would sort of let you see the code that went into making that so far as you can see this is a very simple AI does pretty much what any other AI would do except this has the flash answer but again the novelty will wear off now why I think this is still a really good product because of this so remember in the beginning we talked about the two products they have one is the lat chat and the platform so if you were to click here you would basically see this is more of a development framework where you can fine-tune models create agents and a bunch of other things like function calling and stuff like that and then you can use that with an API with within your products and services however there's an interesting way you can use this within lchat as well and I wanted to show you that so if you go here and if you click on start building it would then bring you to the platform page as you can see on top you have the platform and you also have the chat so like I said there's a bunch of things you can do here obviously the top two are like creating an agent and then fine tuna model also have access to retrieval augmented generation function calling ways of deploying it directly through an API embeddings uh bunch of other things but again that's not why we're here if you click here on agents you'll be brought onto the agent page and the interesting thing is you can create an agent by clicking here in your create an agent page you can choose any of the models that they have let's say we go with the cestal one which is specific one for code and then you can play around with the temperature which controls the randomness you can uh put some system instructions here you can also add a lot of demonstrations which are few shot prompts basically examples like inputs and out puts and then set up your agent in this way right over here you can name the agent I'll go ahead and name it topnotch engineer if you have all of it set up you can hit deploy after deploying you'll see that there is an option called API and L chat API is basically the fact that you can use the agent through an API but if you turn this on It'll ask you the deploy agent to lay chat and you could hit deploy you can go back to lay chat and then you could call that agent right from here by using the at symbol now this I think is pretty interesting because now you can have multiple agents built out and you can be in the middle of any chat and then call one of your specific agents to come and look at some kind of a problem right as you can see now this is hard plugged in here and then you can just ask any question to that top-notch engineer agent I think this is a really interesting use case because this is very much like creating your own GPT in chart GPT or CL out projects but I feel like the possibilities are a little more expanded here because the way you can fine-tune that model so if I were to go back to the platform you'll see that there is a data sets part where you can upload data sets that you have that has information say about your company or your notes and stuff like that and then you can click here and then create a model that's fine-tuned on that information and then you can create an agent on that specific model as that you can choose the model that goes into the agent so you can fine-tune these myal models which then can go into your agent which you can bring into the lay chat by using that at symbol and I think that is a huge game Cher for non- techies to use AI with Advanced functions now obviously that is going to be a tutorial in itself a separate video however I just wanted to show you that this feature is there for so some of you who already understand how to work with these kind of things go ahead and try it out and let me know what you create in the comments below so that's all I have for this video guys hope you found it useful if you did would really appreciate you liking the video and subscribe for more content like this and with that I'll see you on the next one