# How To Use New Google AI Studio (GoogleAI Tutorial) Complete Guide With Tips and Tricks

## Метаданные

- **Канал:** TheAIGRID
- **YouTube:** https://www.youtube.com/watch?v=-W7QdaO1Q3Q
- **Дата:** 22.05.2024
- **Длительность:** 20:28
- **Просмотры:** 21,248
- **Источник:** https://ekstraktznaniy.ru/video/14294

## Описание

How To Use New Google AI Studio (GoogleAI Tutorial) Complete Guide With Tips and Tricks

Join My Private Community - https://www.patreon.com/TheAIGRID
🐤 Follow Me on Twitter https://twitter.com/TheAiGrid
🌐 Checkout My website - https://theaigrid.com/


Links From Todays Video:
https://aistudio.google.com/app/prompts/new_chat

00:47 Three Models
02:01 Create New Prompt
02:37 System Instructions
02:50 Save Prompts
04:10 Structured Prompt
09:45 Video Prompt
14:09 Audio Capabilities
16:18 Model Tuning Guide


Welcome to my channel where i bring you the latest breakthroughs in AI. From deep learning to robotics, i cover it all. My videos offer valuable insights and perspectives that will expand your knowledge and understanding of this rapidly evolving field. Be sure to subscribe and stay updated on my latest videos.

Was there anything i missed?

(For Business Enquiries)  contact@theaigrid.com

#LLM #Largelanguagemodel #chatgpt
#AI
#ArtificialIntelligence
#MachineLearning
#DeepLearning
#Neu

## Транскрипт

### Three Models [0:47]

you can see well there's actually a fourth one but I'll explain that later so there's actually three models here we've got Gemini 1. 0 Pro which is pretty much a base model this is a simple model that is you know used for a variety of different tasks this one has a standard context length you can see that it's got 30,000 tokens there so if you want to use this model for your standard tasks this one is completely fine as you would use other things however Gemini 1. 5 Pro is a lot better in terms of the capabilities and you can see it's got the 1 million token context length that allows you to do a lot more within a context window and in addition the third model that we're going to be using in this video is of course Gemini 1. 5 flash now Gemini 1. 5 flash is rather fascinating because it allows you to have a also a 1 million token context window however the only problem with Gemini 1. 5 flash is that it doesn't have the abilities of 1. 5 Pro Gemini 1. 5 flash is designed to be a very fast model that you can use for certain things and I'll show you some of the use cases later on so one of the things you want to do with Gemini 1. 5 Pro which is the main model that most people are going to want to be using you're click here and you're going

### Create New Prompt [2:01]

to want to click here and you're going create a new prompt now you're going to be prompted with two different options of creating a prompt you've got chat prompt and then you've got structured prompt so I'm just going to go onto chat prompt because this is just going to be what we're using for now and I'm just going to go on a 1. 5 Pro so now we can see here that we have our standard chat interface so with this essentially what you want to do is I'm just going to call this uh the Gemini demo okay so this you don't have to actually change this name but I'm just going to change this for the video purposes I'm just going to call this Gemini demo and then of course right here what I can do is I can do my system instructions so I could say respond in a

### System Instructions [2:37]

pirate themed Manner and this is just for the example but with Gemini 1. 5 Pro what you can do is you can put something in the system instructions and then right here you can click save and this is going to be something that can save

### Save Prompts [2:50]

you time when you want to test out different things because this is a large multimodal model Google knows that what you want to do is you want to be able to test out a lot of things rapidly and what you can do is you can just put something in here which is essentially a way that you want the model to respond so I can say respond in a pirate themed manner in a really upbeat way and then at the top right you can see right here there's this little save button and you just want to click save and then now can you see that it says right now prompt saved so when I go to view all I can see right here my Gemini demo I just click this and then in my system instructions you can see right here respond in a pirate themed manner in a really upbeat way and I can say hello and it should respond in the way that I want the reason you want to do this and you can see it says aoy there mate fine day to be greeting yada y the reason you want to do this is because in future when you are using the multimodal capabilities such as video audio and of course text or image this might save you time if you want to ask a certain questions in a certain way and you want to gauge how the model responds so that is how you use these system instructions and how you can actually save a prompt so now that we don't need to use anymore what I'm going to go ahead and do now is I'm going to ahead and going to do create new prompt then I'm going to do a structured prompt now this is a little bit different but it is not that hard so

### Structured Prompt [4:10]

basically with a structured prompt in the Gemini AI Studio basically what you're doing here is you're just giving the model examples of inputs and outputs that's going to kind of change how the model response is now I tested this quite extensively and it does actually work for example I could put this as the let's just say YouTube the AI grid demo and then I'm going to click save and then what I'm going to do is the optional style instructions for the model I'm going to put you are and right here you want to put what it's going to be acting as going to put you are a senior title writer for a YouTube channel called the AI grid okay and then I'm going to say you convert news into shocking headlines for maximum engagement Okay so this could be something that you want to do now essentially the reason you want to uh do this is because you want to be able to have the model respond in a certain way and I'll show you another way to do this in a moment but this is a very basic way so one of the things you're going to want to do here is in the input you're going to want to change this and then this is where we put titles so this is the classification for our input so if we were putting marketing copy it would be there and then the output would be whatever we want the output to be and we would put the new headline here so for example we could put some examples I could put Microsoft just revealed their new aipc then I could put the new headline is Microsoft shocks the entire industry with new AI PC okay then I could put open AI reveals GPT 5 okay and basically all I'm doing here is I'm inputting the information and then I'm inputting how it should respond and basically what you want to do whilst doing this is you want to take into account these certain bits of information so that you get the response correctly and I'll explain this further but basically what I mean by that is that for example you know how I've just put in that Microsoft just revealed their new aipc what it's going to do is it's going to look at the key text here so Microsoft and PC and it's going to see how I've taken it from there and input it into there so as you can see Microsoft shocks the entire industry I'm going to put open AI shocks the entire industry with new GPT 5 model and then I could put uh anthropics reveals state of the art clae 4 and it surpasses GPT 5 then I could put um in here and this is what I want the kind of headline to be so on the right hand side here this is exactly where I want or what I want my response to be like so you can see right here I put anthropic reveal state-ofthe-art claw 4 and it surpasses GPT 5 anthropic claw 4 just done the entire industry and basically I'm just using some examples now here's where you get to testing your prompt so if you've written these what you want to do is you want to then test your prompt with the information and you want to see if it actually works so I going to put uh let's see what other companies are Google has just revealed their new Gemini 2 model and it's it surpasses GPT 5 so now then I'm going to put uh run output and then this right here and I click this button so I click generate response and then I'm going to see it says Google Just Dropped a Bomb Gemini 2 crushes GPT 5 so this is how it works and you can kind of test this if you want to by removing these examples so if I remove these examples I'm just going to go ahead and remove these I'm going to remove these examples and then you can see it said Google Just Dropped a Bomb Gemini 2 crosses GPT 5 and then I'm going to just go ahead and run this again we're going to see that um it does your titles in a very different way so it does your titles that you might not expect but the point here is that what this actually enables you to do is it enables you to format information in a very specific way that you might want to use I think this is only useful if you have a very specific style of you know piece of information a very specific output and then you want to be able to use that in the future to be able to get that consistent kind of output and whilst yes you can do this with standard chat interfaces the reason you're going to use this is because it covers a very different amount of examples that you're going toward to use so for example one of the examples that Google actually did use is they did this so you can see right here it says Identify all brand names mentioned in the input multiple products will be separated by commas so on the left hand side we can see just a standard piece of text and then on the right hand side you can see it outputs all of the brand names now the reason that this is good and actually works is because it has more examples so the more examples you use the better it's going to be and what we can also see here is that the model is Gemini 1. 5 flash because the output is fairly small and you want to use Gemini 1. 5 flash if you're not entirely writing new content this is literally just taking out the content from here like Tesla Google and Disney and you can see it just writes out the username and then we can see here that we can literally just click run right here and it's going to give us the exact outputs that we need for this task so this is something and an example of where you want a very basic task done at scale and you need a few examples this is exactly how you can utilize the model to I guess you could say use it for certain use cases and if you want more use cases I'll have another video perhaps attached to this one or a link in the description that's going to show you guys all the different ways that you can use this now with 1 Point 5 Pro of course one of the

### Video Prompt [9:45]

key features is the context length so what you can also do is you can test how good this is so one of the things that you might want to test is one of these sample videos so when you are here you can see that there are these sample videos and you can literally ask Gemini what this video is about so you can say what is this video about and whil yes this does take some time because it is a 10 minute long video and we can see on the right hand side the token length that it is 177,000 tokens and then we can see how long this takes to respond another thing as well is that something that most people don't realize is that Gemini 1. 5 flash is also multimodal just a lot faster it's something that's used for different tasks but with 1. 5 Pro this is of course the big task SL big model that you want to use for the majority of your long context queries so here you can see we get our response and this response is pretty small now I'm guessing what we could do is if we wanted to we could also click rerun edit right here so we could edit this response and we can edit that based on the entire thing that we want and another thing that you should know is that often for whatever reason the responses in this system are not that long so I would say since the responses are not that long it's kind of best to ask Gemini 1. 5 Pro to make its responses as long as possible that's just a tip that I've noticed when working with videos and audios because it usually just gives a very short description which is much shorter than it would do if it was in a standard chat user interface I'm not entirely sure why but if you do want a lot more information you can see right here I was thinking that it was going to give a you know detailed information but since I just said what is this video about it doesn't give me that much now you can see right here that I've asked it for a detailed explanation as summary of what you see in the video and it doesn't do that much of a great job now I think what Gemini actually does excel at and what they showed us in the actual use cases video where they showed us a few demos is that Gemini actually excels at being able to identify specific things that go on at specific times so if you ask a Google Gemini about a video and you ask it for something that happens at a specific time then that is something that you can use this for because it's able to identify all of the different frames and then give you a time stamp for what happens there and then able to give you further context on that it's not something that you want to use kind of for like video summaries but it's more of like a tool that you can use to find things in really long videos so if you do want to upload a video yourself you can just click your drive and then right here you just want to click upload and that's how you upload your files so what I'm uploading is a recent thing from Microsoft's demo about their new co-pilot software and I'm just going to ask a couple of questions about that demo cuz I want to know exactly where certain things appear so I'm going to say at what time stamp do the zombies appear and then I'm going to click run so I'm just going to wait for it to tell me when the zombies appear so it says that the zombies appear at the time stamp 59 seconds so I'm going to go ahead and check that and here we have the video so then I'm going to go ahead and all the way to 59 seconds and then at 59 seconds the zombies just start to appear so this I guess you could say is pretty right they don't appear exactly at 59s seconds they appear like 2 seconds later but this is actually the time stamp you want to be if you do want to find the correct thing and this is basically just another video that you've probably already seen on my channel but the point here is that this is something that can find specific things in certain videos and it's able to give you more context on that then you can see here I asked them what the video about and it says this video is about a person playing Minecraft getting assistance from a Microsoft AI assistant and it's able to provide helpful tips yada y yada so you can see that Gemini 1. 5 Pro is able to do long context on standing across videos which is really useful in addition you could always use the sample videos to test the full movie of CC Julia from 1924 this 30 minute thing but I would say that you know it's going to take a longer time to respond because there is more footage and another thing as well is that if you do open up a new chat right here you can see that you don't actually get to save these chats so if you do want to somehow save your interactions just click the top right hand there to click save and then that'll save everything in the chat so I'm going to go ahead and use a new chat prompt to show you guys some different use cases so here I'm going to show you guys how to actually use the audio capabilities so this is a video script that is basically project

### Audio Capabilities [14:09]

Stargate if you don't know it's a superc Compu that is going to be used to build AGI you can see that this is actually around 16 minutes long it's got 32,000 tokens and I'm going to ask it something about this audio so what is this audio about and then we're going to see how long this takes now what you can actually do is if you want something that's pretty quick and I wouldn't recommend this you can use Gemini 1. 5 flash a 1. 5 flash is a model that is much faster than Gemini 1. 5 Pro so I'm going to show you Gemini 1. 5 Pro first so I'm going to say what is this audio about and we're going to wait to see how long that takes and we can see right here that in literally around 10 seconds it was able to give me this incredible response which is much longer than the other one so this seems to be really comprehensive in terms of its audio you know capabilities it being able to summarize exactly what's going on so audio capabilities are really cool so what I'm going to do now is I'm going to just do the same with flash so what is this audio about and maybe you'll see a marginal difference so I'm just going to run this and we're going to see how quickly it's able to respond because 1. 5 flash should usually be yeah you can see 1. 5 flash is so quicker in how it responds you can see 1. 5 flash also does have shorter responses you can see it talks about all of the phases of project Stargate so it's pretty similar it just also costs a lot less than the 1. 5 Pro so if you're wondering what I should be using in my kind of project 1. 5 flash is something that you use for quick image identification quick audio recognition maybe you want to classify what the audio is rather than get detailed descriptions that are really necessary and you need to save money that's where flash comes in you can easily identify an image saying you know it's a bird it's a plane this is an audio about you know for example AI this is a video about AGI ASI but if you need really specific descriptions that's when you go to Gemini 1. 5 Pro and that's when you ask it certain questions that actually take advantage of its much more capable system so now something actually really

### Model Tuning Guide [16:18]

cool that you can do with Gemini is that you can actually tune a model so right here you can see you can click this button called new tuned model so of course click okay then you can see right here we can then tune the model so this is where you can basically have a model that is tuned on your data that's going to perform a lot better and adhere to your tasks in a certain way so right here it says select data for tuning you can tune a model from an existing prompt structure or create one by importing Google Sheets or a CSV which is basically just an Excel file tuning only works with text at this time and we recommend using 100 to 500 examples there the entire tuning guide but this is what I'm here for so I would recommend not using one of the structured prompts cuz it's just a little bit confusing I would just choose an existing prompt and right here and then of course what you want to do is you want to click import now what you can do is this is what I did because I wanted to actually test this out well I actually just asked chat GPT to literally generate the data for the tuning just to see how it works and basically if you don't know how the data should be structured so here you can see I asked it to do 20 examples for me cuz 20 examples is just the base but you can ask it to do 8 100 examples or you can just ask it to do 20 examples batch by batch and it will do it and then you just click this button to download it then this is what you can see in Google Sheets or Excel whichever software you're using and then this is uh what you'll have the question and then the answer so it's usually this is the format it's just like input and then response and then this is exactly what you'll need so just double check this to make sure that this is right and then uh you'll know that this is good okay so then you just want to click import right here then you just click insert and then you just wait right here then you can see right here that this is what's coming and then right here you can see you just want to do new input column then right here just new output colum then of course just prefix just leave these both ticked then you just want to click import 20 training examples and then you can see right here preview showing one of 20 and you'll know that is currently working now you don't need to go ahead and change uh any of these advanced settings usually they'll be completely fine these advanced settings are only if you really want to just change things and just really experiment with stuff but this is uh you know like it will automatically update based on the number of training examples that you have like for example it will change this to 0. 01 based on the fact that I only have 20 examples because uh you just need like a lower number if you want it to work more effectively so this is the tuned model name and I'm just going to put Gemini tuned model demo and then we just want to click here tune so once we click here tune that should be going and then essentially what we do now is since we just did that we'll just wait for it to pop up so you can see right here it is now in the queue and then we're just going ahead and click this to wait now basically it has this graph and usually what will happen is the graph will stop or where it will Plateau should be around the end of the epoch if it doesn't then that means something's kind of wrong and you just got to do it again but essentially you can see right here that the OC is around 5 which is pretty good so you can see if you want to train this again you can actually just stop it at 4 because it plateaus at around 4 there's no like further Improvement after 4 so you could literally just do that again and stop it at 4 if you want it to go a bit quicker and of course you can experiment with other stuff but now that we have our tuned model based on whatever thing that we want you can see right here we can see C tuned model or we go into a new chat and then of course for our new chat we want to click view all and then Gemini tuned model and then it says use your tune model just use instructed prompt and of course now we can us just use that based on the training examples that we do have there and this is exactly how we use Gemini 1. 0 to be trained on specific examples now if this tutorial did help you with how to actually use the software please let me know if there are any questions that you do have about any of this stuff I will try my best to answer this if there are any bugs and stuff don't forget to use the comment section below as a place to share data and questions and comments but after that I will see you guys in the next video