Copilot Pro: Custom GPT Builder

one of the big Promises of the co-pilot Pro licensed announced in January alongside faster GPT 4 access and the addition of co-pilot in the Microsoft 365 apps if you're appropriately licensed was the addition of a custom GPT Builder reminiscent of what we've had as part of chat GPT Plus for many months that custom GPT Builder is now here to preview so in this video I'm going to take a look at it what can it do how does it compare to the GPT build Builder that's part of chat GPT and how does it fit in to The Wider co-pilot landscape as always all the screen recording you'll see in this video is done in demo environments where you're not seeing anyone's private information but before we dive in a quick introduction my name is Nick deori I'm the owner of bright ideas agency a digital transformation consulting company focused on the needs of small and mediumsized businesses I share information here to help you navigate the artificial intelligence age with a particular focus on Microsoft 365 and co-pilot if you enjoy what you see here then it would be great if you'd give the video a like to help you get in front of more people and please subscribe to the channel if you want to keep up to speed on topics such as this so jumping into Coop pilot. microsoft.com with a co-pilot Pro enabled account we can now see you get co-pilot gpts on the right above your chat history by default you're seeing a list of Microsoft's created gpts designer vacation planner cooking assist assistant and fitness trainer as well as a link to see all copilot gpts if you click on this the top option is to use the GPT Builder preview but I suppose before we dive too far in here there's a question many of you might be asking what is a GPT GPT stands for generative pre-train Transformer and this is really the type of AI model that's in use here and while we generally use the term large language model to describe what the GP T4 model that underlies co-pilot is it is more accurate to describe it as a GPT model this makes the term custom GPT confusing as this would imply you are building a customized AI model but this isn't the case nor is it the case when you build a co-pilot with co-pilot Studio the underlying AI tooling Microsoft provides is the same I think it's easier to think of it from a branding perspective we're familiar with chat GPT as a brand but other AI services like Auto GPT also take this naming approach to highlight a specific tool and its connection to the GPT model when you build a custom GPT whether you're doing so with co-pilot or with chat GPT what you're doing is giving your tool some form of bounded contextual data and additional system prompts that control how the AI model interprets and responds to the requests that are made while these techniques are useful to customize your experience they are very different to building a custom AI model from scratch for a purely technical perspective understanding this difference is important but as we explore these capabilities you'll see that understanding the Practical implications is also vital so if we click on the button to create a GPT you are presented with a co-pilot GPT settings screen and instructions on how to get started from here you can converse with co-pilot to build the GPT and upload files related to what you are trying to build you can also jump over to the configur tab where you can see what is going on in the background and you can define a name a description create instructions upload knowledge and decide what capabilities the GPT should have either web browsing or image generation I'm going to start with something really basic I'm going to create a copilot GPT to help answer questions about Microsoft 365 and I can do that with natural language just by prompting copilot I start with a prompt you will be an adviser for users of Microsoft 365 answering questions about how to use the apps licensing and other related questions related to the Technologies in the Microsoft 365 Suite it processes that request and next it suggests a name for your co-pilot GPT you go through a conversation where it asks for clarification around how to answer questions or any special requests but from here you are just about ready to roll you can continue to Tinker here if you want though in the background on the configure page your conversation is being turned into instructions so Copart is working with you to essentially tailor itself to your needs you can review the instructions and work with them here or return to the chat and continue to work conversationally but when you are ready you can preview your co-pilot GPT and then you get a co-pilot window with the name of your co-pilot GPT at the top and you can just go ahead and use it like you would co-pilot normally in this case asking questions about Microsoft 365 and getting perfectly adequate responses so you might be thinking what's the point point of this I could just ask co-pilot about Microsoft 365 normally and that's true this is a basic example but the more nich or specific your need the better this approach can help as you can work to tailor the type of response you get for example building a tailored approach to how the response is constructed so it's more useful to the way you research or the types of writing tasks you do if you're happy with your copilot GPT you can go ahead and publish it either for you or as a shared GP T and then you could use it along with other gpts you have however the power of this approach is not just about customizing the way inputs are interpreted or outputs are generated it's also about bounding the context data co-pilot uses to service the request let's try another slightly more complex example this is the Power Platform licensing guide for February 2024 it's a fairly heavy 33 page PDF that details everything one might want to know about platform licensing and until Microsoft releases a later version this is what you should rely on getting tools like co-pilot or chat GPT to respond with this level of specificity in where answers come from is ordinarily pretty hard and you end up saying you want to use the latest information but find you get a web page or 5 years ago included in the result here I'm building another GPT but different to last time I'm going to upload the Microsoft Power Platform licensing guide and be highly specific about the fact that I want the answers to come from that resource there are any number of potential use cases where strictly bounding the context of data will be important to efficacy perhaps you are building a legal GPT and you want to ensure the law documents used are from your country or you're a fan of a particular public domain book series and you want a GPT that is going to answer questions addressed by the author in their work but not the analysis of other readers from across the internet so I jump into the GPT to preview it and ask a question that definitely requires some fairly current information how do I license co-pilot studio and how much does it cost and it gives me an answer but despite me being specific about what source I wanted to use you can see it referencing websites and while Microsoft sites are likely to be equally up toate I'm not sure about trust radius as an upto-date source so I don't want my GPT search in the web at all I want it to use the document I uploaded as its knowledge so I could jump back into my configuration screen for the GPT turn off web browsing and try again and now it ignores the document I uploaded completely and tells me it doesn't have any information as of its last update so what's going on here I took time earlier in the video to focus on the fact that despite these being called gpts you are not building a custom AI model you are just building a set of custom instructions sitting on top of the existing AI model gp4 by uploading a document you're using a technique called rag or retrieval augmented generation this is where you essentially interject additional contextual information into the cycle that you're getting of the request and the response and by using that additional contextually bounded information you end up with a better response this is essentially how co-pilot for Microsoft 365 works p pt4 is still in there and still has the limitation of having been trained on a vast set of data with a cut off point well before co-pilot Studio was ever announced or released and despite me uploading appropriate contextual information and being specific that this is what I want co-pilot to use it relied upon its internal training knowledge to answer my question rather than the document so it didn't actually use that rag technique at all I can get it to use a document by prepending ending the phrase using your uploading knowledge to the request but given that I could upload a document normally and do this it seems like a strange workaround but in this case it gives an answer that clearly comes from the document the problem here though is that in this specific case it's very clear that the document wasn't being used to answer the question but this is kind of a fluke based on me asking for something that happened after the end of training GPT 4 there are an endless SE possibilities of documents I could upload connected with topics that are in GPT 4's training data set and I could be inadvertently getting answers I would think would be coming from my document that are instead coming from some unknown source gp4 was trained on but I guess this is a hard problem to solve right let's take a look at chat GPT to find out but before we do I want to ask you a question what's your approach to AI adoption in your business if you're yet to start building your plan and you're not sure where to start you might benefit from my new on demand Course fly into the age of co-pilots focused on helping leaders in small and medium-sized businesses learn about AI technology and its benefits and right now for a limited period it's completely free check it out at the link below so here I am in chat gpt's GPT Builder and I've set up a GPT with exactly the same instructions file for knowledge and settings such as web browsing being turned off that I had in copon pilot and I'll try the same prompt chat PT is successful in using the file straight out of the gate but interestingly it does give an answer that is less nuanced than the one co-pilot gave when you forced it to use the file it details co-pilot Studio licensing but does not differentiate the fact that co-pilot studio is also available to co-pilot for Microsoft 365 users in the way the co-pilot did if I turn web browsing back on I get much the same answer and there's still no evidence that chat GPT is searching the web for more context this answer is a little broader but off first review it appears it's all information that would have come from the licensing guide on this simple test I'm not impressed with co-pilot's GPT Builder I think wanting to apply boundaries to the context is a fairly foundational use case for this sort of tool this rag capability is pretty core to the value proposition of this sort of product and co-pilot just doesn't seem to be as good at it as chat GPT even if when you kol co-pilot it definitely uses your context to provide a moderately better answer additionally there are just more options in chat gpt's GPT Builder to help you build something more sophisticated by turning on code interpreter you get access to features like creating files with ch GPT and the actions option gives you the ability to call any external API to get data from exactly where you want with up to the minute accuracy frankly the GPT builder in chat GPT has a set of features that align somewhere as a Middle Road between co-pilot's GPT Builder and co-pilot studio and in this I think Microsoft has taken a bit of a wrong turn here assuming the GPT Builder works as described it's in preview right now and it clearly needs a lot of work before it's out of preview I could see a lot of business users with co-pilot for Microsoft 365 licenses who would be completely scared off by the sheer complexity of co-pilot Studio who might find Value in a basic interface like the co-pilot GPT Builder or the chat GPT GPT Builder to create small enhancements to the usability of co-pilot for their day-to-day work imagine if you could turn on different gpts in Outlook to draft emails for different purposes perhaps with a different set of knowledge documents and share them around your team in an interface as straightforward as the co-pilot GPT Builder you could probably teach an entire Team how to do this end to end in the same amount of time it would take to explain all the options on the main menu of co-pilot Studio but conversely those who are passionate enough about AI to Shell out $20 per month on their own co-pilot Pro subscription are often probably the same people who are excited to try out features like chat gpt's API actions and would love to get their hands on the power of co-pilot studio if you're a more basic user with no interest in playing with AI then you're not going to buy co-pilot Pro you're not even going to try co-pilot unless your business buys you a license are you a co-pilot Pro user do you agree with this point or are you pleased to see a basic GPT Builder come to co-pilot let me know down in the comments overall I appreciate that Microsoft is bringing this capability that seems to be so liked by chat GPT plus subscribers to co-pilot Pro they need to make it work better and for equivalency they need to add a couple more features but if copilot pro has a GPT Builder on par with chat gpts then copilot Pro's value becomes even more compelling especially for those who use or are interested in using the Microsoft 365 productivity suite for personal or family needs however again this is a case of a product clearly marked as a preview but with big issues that need to be fixed I have no issue with there being previews that are somewhat broken but the degree to which Microsoft continues to eliminate the friction of getting Hands-On with preview features alarms me it has been the case for decades that those who are technically Adept or just adventurous could try out pre-release features across lots of software but there was always a level of friction perhaps you needed to download a completely different install package for example that made it clear you were doing something that was out of the ordinary but then came Google's Gmail that was in beta for about a decade and all sense of users actually understanding the care that should be used in relying on pre-release software vanished and with co-pilot and frankly across Microsoft 365 Microsoft continues with this trend with any of these previews is it actually clear what features are being previewed and which we should be cautious of is it just the concept of the software overall is our investment in building gpts likely to be for nothing as at the end of the preview they'll all be wiped and we'll need to start again we are also used to using software where it's entirely unclear what should be working and what shouldn't and which features we should be able to see and which we shouldn't and there's always the likelihood that I record a video like this and a few days later when you see it the system works in entirely differently without any indication that anything has changed anyhow there is one good thing about all these changes lots of content for new videos thanks for watching through to the end until the next video bye-bye

Channel: Bright Ideas Agency

Views: 3,629

Keywords: microsoft copilot tutorial, how to use chatgpt 4, copilot pro, microsoft copilot pro, copilot gpts, copilot gpt, copilot gpt builder, copilot pro gpts, chatgpt, chatgpt plus, copilot studio, ai explained, artificial intelligence, copilot pro tutorial, microsoft copilot pro tutorial, microsoft 365, copilot for microsoft 365, microsoft 365 copilot, chatgpt tutorial 2024, ai automation

Id: KzQDXeaFRec

Length: 16min 29sec (989 seconds)

Published: Thu Mar 21 2024