Live Demo of ChatGPT Canvas & Overview of Microsoft Copilot’s Wave 2 [Sidecar Sync Episode 51]

Timestamps:

00:00 - Introduction
02:15 - Hurricane Update & AI Weather Forecasting
05:58 - ChatGPT Canvas
08:00 - Live Demo of ChatGPT Canvas Features
13:42 - Further Applications
27:00 - Content Transformation: Quality vs. Quantity
34:16 - Microsoft Copilot’s Wave 2 Highlights
32:45 - Content Strategy for Associations
43:13 - The Future for AI Tools
52:40 - Closing 

Summary:

In this week’s episode of Sidecar Sync, hosts Amith and Mallory dive into two exciting AI innovations: ChatGPT Canvas and Microsoft Copilot’s Wave 2 updates. They break down how these tools can enhance productivity and transform workflows in associations and beyond. From the advanced features of ChatGPT Canvas, enabling real-time edits and content creation, to Microsoft Copilot’s newest collaborative capabilities, including Excel’s Python integration and Copilot Pages, this episode provides actionable insights for AI-powered productivity. Whether you're exploring AI for content transformation or team collaboration, this episode covers it all!

Let us know what you think about the podcast! Drop your questions or comments in the Sidecar community.

This episode is brought to you by digitalNow 2024, the most forward-thinking conference for top association leaders, bringing Silicon Valley and executive-level content to the association space. 

Follow Sidecar on LinkedIn

🛠 AI Tools and Resources Mentioned in This Episode:

ChatGPT Canvas ➡ https://openai.com
Microsoft Copilot ➡ https://www.microsoft.com/en-us/microsoft-365/copilot
Anthropic Claude ➡ https://www.anthropic.com
Replit ➡ https://replit.com
Cursor ➡ https://www.cursor.com
Sidecar Sync Ep. 48: Unlocking AI-Powered Insights from Unstructured Data ➡ https://youtu.be/QyPV7A6VRn4?si=cZMnmFNYeE3Lkcew

More about Your Hosts:

Amith Nagarajan is the Chairman of Blue Cypress 🔗 https://BlueCypress.io, a family of purpose-driven companies and proud practitioners of Conscious Capitalism. The Blue Cypress companies focus on helping associations, non-profits, and other purpose-driven organizations achieve long-term success. Amith is also an active early-stage investor in B2B SaaS companies. He’s had the good fortune of nearly three decades of success as an entrepreneur and enjoys helping others in their journey. Follow Amith on LinkedIn.

Mallory Mejias is the Director of Content and Learning at Sidecar, and she's passionate about creating opportunities for association professionals to learn, grow, and better serve their members using artificial intelligence. She enjoys blending creativity and innovation to produce fresh, meaningful content for the association space. Follow Mallory on LinkedIn.

Read the Transcript

Amith Nagarajan: Greetings, and welcome to the Sidecar Sync, where you can get all the information you need about associations and artificial intelligence, and how AI can be applied to moving your association forward. My name is Amith Nagarajan.

Mallory Mejias: And my name is Mallory Mejias.

Amith Nagarajan: And we are your hosts before we get going on an exciting episode where we're going to be talking about a couple of really cool new AI tools and how they apply to the world of associations.

Let's take a moment to hear a quick word from our sponsor.

Mallory Mejias: Amith, how are you doing this lovely Wednesday morning?

Amith Nagarajan: I'm doing great. It's, uh, weather here in New Orleans is actually quite nice and, uh, enjoying that. So it's great. And, uh, Thinking a lot about the folks in Florida who are, uh, you know, bullseye for this new hurricane that's in the gulf at the moment. So hoping that that is, uh, somehow, you know, dies down a little bit before it hits them.

Mallory Mejias: for sure. Yeah, our hearts go out to everyone in Florida who just got impacted by one hurricane and are now facing another. It's really, it's, it's devastating for sure.

Amith Nagarajan: You know, I think we've talked in the past about, uh, a couple different times about A. I. Weather forecasting and models. I think in the world of hurricane prediction. Um, one of the good things is, is in the last I don't even the last probably the last 10- 15 years, the accuracy of the forecasts have already been really, really good.

Part of what's not easy to forecast is, um, you know, the speed at which the hurricane is going to come in and storm surge and things like that. But they've gotten pretty good at determining the path of the hurricane. So, you know, It's much less likely to be like, Oh, well, we think it's going to hit Tampa.

And then it ends up hitting Pensacola or something instead. Um, so that's good. But I think as these things become more and more accurate to say, Hey, like three weeks from now, we're going to have another hurricane come the more time people have to plan, uh, the better, obviously with these kinds of events.

Mallory Mejias: Yep, I mean obviously this is not our area of expertise by any means, but it sounds like this is an area where AI will continue to see improvements in terms of predictions. And I'm assuming there's probably lots of artificial intelligence already being used, Amith, to predict these things. Is

Amith Nagarajan: Well, the current models that are in production, both in Europe and North America are numerical methods that, uh, predate AI. So they're not actually using AI methods for. The National Hurricane Center for the European models. And that's for like the general weather forecasting. Um, so it's, it's very high compute intensive numerical methods that are actually really good and really accurate, but they're super, super expensive.

And their forecasting is at this very, very high level in terms of granularity. So part of what AI allows you to do, and we've, we've talked a little bit about this, but. Both with the deep mind work and also with Microsoft's model, which is more atmospheric composition stuff, um, is that you're able to get really granular and have near instant, um, forecasts at the very micro level.

So being able to say what's gonna happen in this zip code or this postal code. So I think that's gonna be an interesting innovation. So the two may complement each other. Um, at some point, the A. I. Models might supersede the traditional models, but that's hard to say right now. But I'm really excited about it because I think Further out you can look, the better and the more specific you can get, uh, the more it helps people because, you know, what happened in Asheville, obviously, uh, was terrible and very difficult to predict because you wouldn't have thought that up in the hills that you'd have those kinds of, of results, but, you know.

So, It's a combination of the soil conditions and the, uh, the how long the hurricane was over that area and how much precipitation had dropped, obviously, that caused, uh, such terrible damage there.

Mallory Mejias: Yeah, for sure. It was, it was very eye opening for me because I often, I don't write off bad weather, but it's not something I look at every day unless I feel like there's going to be a really bad storm or, you know, or tornado or something like that. But I think you make a great point with Asheville that it just seemed so unexpected.

And so having that technology that could even predict some percent likelihood that that would have happened, I mean, would have been very helpful.

Well, we are thinking about everyone in the hurricane's path in Florida and wishing everyone the best.

Today, in episode 51, we have two exciting topics lined up for you all. The first of those will be ChatGPT Canvas, and the following topic will be Microsoft CoPilot's Wave 2. ChatGPT Canvas is a new interface launched by OpenAI this month to enhance user interactions with ChatGPT for writing and coding projects. Canvas opens a separate window alongside the main ChatGPT window, allowing users to work on writing and coding tasks more efficiently. The setup enables users to generate, Edit and refine content in a dedicated workspace.

If this sounds slightly familiar to you, it's because Anthropic's Claude has a similar feature called artifacts that we've covered in this podcast before, but I will dive into the, how they're a little bit different from one another. Within ChatGPT Canvas, users can highlight specific sections of text or code to receive targeted edits or suggestions from ChatGPT, making it that much easier to iterate and improve final output.

Canvas offers several features to improve written content as well that are built in, and I'm going to share my screen in just a bit to show you these, but we've got length adjustments, so users can shorten or lengthen their documents, reading level adjustments, A final polish, where ChatGPT can refine text for grammar, clarity, and consistency.

And emoji formatting, so a basic emoji insertion are supported as well. With coding, we've got code review, comment edition, which is automatic generation of inline documentation, identification as well. and fixing of bugs and conversion of code between different programming languages. ChatGPT Canvas is currently available in beta for ChatGPT Plus and Teams users with plans to extend access to Enterprise and EDU users.

OpenAI intends to make Canvas available to all ChatGPT users once it exits the beta phase. To access Canvas, you can select GPT 4. 0 with Canvas if you have one of the accounts that we just mentioned. Ask Chachubiti to use Canvas, which will trigger it, or it automatically detects when it thinks a canvas would be beneficial to you. One second. We'll edit out this pause.

So right now I am sharing my screen to show you all what canvas looks like within ChatGPT. I used the transcript from last week's episode actually where we talked about our brand new and improved AI learning hub. I asked ChatGPT to act as an expert copywriter who is skilled in writing copy that converts.

And then I asked it to write a blog about our new AI look. AI Learning Hub launch based on that portion of our podcast transcript from last week. As I did this, it automatically popped up this canvas window. So you can see it generated a blog, um, A pretty decent sized blog, and reading through this just at a glance, I would say this is pretty good.

There's a couple of subsections, a new approach to AI education for association, new features like our AAP certification, and the community. Why AI matters now and are you ready to lead your association into the future is kind of the cta part of that blog What's really neat about this is you can actually make edits in the canvas Which is not something that you can do with claude at this moment I imagine that they will be changing that really soon based on this release of canvas so There's a sentence here that says we've moved to a better learning management system to enhance your You Experience.

I can actually go in here and say, well, I don't really like the word experience. Maybe I want to say learning journey instead, and I can make edits directly here, which is nice. So you can say goodbye to the days of having to generate a whole blog or social post at once, take it out, put it into a Word doc, and make your edits there.

You can actually make all your edits right here within ChatGPT. Another interesting feature that I will point out is the ability to highlight a specific section of text and then make an edit just based on that. So right now I'm working with that same sentence that says we've moved to a better learning management system to enhance your learning journey.

I think maybe our audience might be curious on why we did that. So I'm actually going to highlight this section of text. And say, add an explanation on why we moved to a new LMS. And you can see right here in Canvas, it edited just that portion of the blog. And then finally, I want to point out these, uh, built in writing features that I discussed earlier.

So one of those is adjusting the length. If you click this button in the bottom right hand corner, you'll see there's actually this drag tool where I can make it longer, longest, shorter, shortest. So I'm going to make it a little bit longer just to see what happens. And we see that it's rewriting the blog to be even more wordy.

I will also demo that you can change the reading level within Canvas as well, which is something that we're going to discuss in terms of transforming content that you already have. Wow, it made this significantly longer. So now I want to adjust the reading level. The features that it has built in right now are high school reading level, college, graduate school, middle school, and kindergarten.

So I'm going to up it to, let's say we're writing this for folks in graduate school, just to see what this transformation looks like.

It's not all that different. I'm curious to hear what you think, Amith, on this, but you can see, it's a bit wordier, maybe a bit more verbose. What do you think?

Amith Nagarajan: I'd love to see the kindergarten version. If you want to change to that,

Mallory Mejias: that's what I was thinking. Maybe we can kindergartenify it. And

Amith Nagarajan: more. And honestly, it might be a lot better because, you know, a lot of times the simpler, the copy can be the better. So let's see what happens here.

Mallory Mejias: exactly. Let's, I don't think this has ever been done. Let's talk about the AI learning hub for kindergartners. Who, debatably, couldn't even read this. So, that's kind of interesting. Sidecar, okay, AI, or artificial intelligence, is growing and changing fast. And Sidecar wants to help you learn about it. Mmm. I don't know that I would say this is quite kindergarten level.

Amith, what do you

Amith Nagarajan: Yeah, it's probably a little bit above that, but it's interesting because what it did is it actually summarized it and made it a lot shorter. That's the main thing I can see from a quick scan of it.

Mallory Mejias: And it seems like some of the words are really simple, fast, easy. Um, I'll also point out, I haven't done this one yet, that you can add emojis to whatever you're working on in Canvas. We do like to use emojis here at Sidecar. Ooh, okay. This is a little bit

Amith Nagarajan: That's a lot of emojis

right there.

Mallory Mejias: There's an emoji or more than one in every single sentence.

It looks like. But that is Canvas. So as I said, you can make edits right here. You can also continue working in the regular chat interface as you're used to with ChatGPT. But, to me, being able to make these inline edits is huge. Really, it's a huge time saver, but it's also huge in terms of the quality of output that you're going to get when you're using ChatGPT, because you can actually infuse your own expertise and your own style into what you create.

So that was a quick little demo of ChatGPT Canvas. Now, Amith, I'm going to quote you really quickly. You messaged me on Teams about this and you said, ChatGPT 4. 0 Canvas Beta is ridiculously well done and lightning fast. So I'm going to ask you to back up that quote with our listeners and viewers.

Amith Nagarajan: Sure. Well, first of all, for those of us, uh, for those of you that are only listening, uh, you might want to check us out on YouTube because that's where you can see the full videos and see the demos we're going to be doing more and more of this over time where we have. Interesting things to show. Um, so as far as why I was impressed by it, I thought they just really did a good job with the software engineering of creating an application that was easy to use.

Pretty intuitive. Um, I liked the in line editing capability. That was really cool. And I also like that. You can highlight things and tell it to change just that one portion. So, you know, It's like interacting with another person where you can say, Hey Mallory, I really liked that blog you wrote, but this one piece, I'd like you to change it in this way or that way.

It's, it's more of an approximation of how you'd collaborate with another human, as opposed to the way we've been dealing with AI, as amazing as this AI technology has been in creating copy or creating code, uh, generally speaking, each time you asked for a change, you would rewrite the whole thing. And so you'd have these monstrously long chats with Claude or chat GPT prior to artifacts and Claude and prior to canvas and chat GPT, where you would see the same thing over and over and over again, it became super repetitive. Um, so it's hard to find the change and hard to actually work with the tool quite as fluidly as you'd like. So to me, this is not a change in the underlying AI model at all.

This is just more of the software engineering on top of the model that makes it more effective, makes it more easy to use. Essentially. So I was blown away by it for that reason. And I think there's so much upside. This is a really good illustration of something we've talked about a lot on this pod, which is that the state of the art in terms of what models can do has barely that we have barely scratched the surface in terms of the applications we can create using current AI. So, you know, here's a thought experiment for you.

Let's just say hypothetically AI just froze that there was no improvement beyond four Oh and Oh, one caliber models. Things just stopped and there was no improvement at all for, say, the next five years or the next 10 years in a I very unlikely for that to happen, obviously. But let's just say it did in that scenario.

Let's say software engineering, though, was, you know, going full blast ahead, saying, What can we do with these models? There would be innovation after innovation after innovation, where we'd be stacking up capabilities on top of one on top of the next on top of the next, creating new applications for use.

Um, it's kind of like when the mobile phone first became a big thing in the context of, I should say, smartphones like the iPhone and its earliest versions, even apps, you know, products before that or when Android came out. Um, it was a platform that needed applications and some of the first applications were games that took off.

Then after that, people found use for productivity, but it was a platform in need of applications. And in some ways, um, AI is a new paradigm for computing and you need to have applications on top of it that are actually useful for end users to take advantage of. So what I loved about this is that this is a really big step forward in making these tools just so much more helpful in day to day work for for typical end users. It's more intuitive because you've gotten used to editing in line and word and and every other app. So of course, you expect the I to work that way. So to me, that was what was exciting. Chat GPT is dramatically bigger than Claude in terms of its adoption in the user community.

I think it's like eight to one or something like that in terms of the number of users and chat GPT versus Claude. Uh, and so I think this will have a big impact on that user community. Um, I think Claude's a fantastic product by the way. I know you use it, um, probably as much or more than you use chat GPT now, or have you switched more over?

Mallory Mejias: I pretty much exclusively use Claude at this point, but this might be enough to take me back to chat. GPT with the inline edits. That's very helpful.

Amith Nagarajan: I think that this canvas feature is better executed than Claude's artifacts, but artifacts is great as well. Uh, and certainly it's, it shows how hyper competitive this space is. We're two of the leading AI companies are innovating this quickly in the market. Um, that's exciting as well, because we're going to see more and more benefit.

Ultimately competition benefits the end user of the technology tremendously. So, yeah. Um, I think it's exciting for all those reasons. So for me, um, I I've been using it quite a bit since it came out. I found for anything that's complex. If I'm just having a quick chat with the A. I about some topic I want to learn more about or whatever.

I don't really need canvas, but let's say I'm working on, you know, um, I don't know. Let's just think about like a technical design for some new software application that we're cooking up. Um, and in that process we're kind of iterating back and forth on, you know, what should this design look like? How should it work?

What's the database look like? What's the, what does the business, uh, workflow look like? Those are all like technical details that you might have in a, in a document or in some kind of format. And you want the AI to be able to work in context to say, oh, okay, no, that part of the database design isn't quite right.

Change it this way. And up until this type of UI, you have to go and basically rewrite the whole thing over and over, which was really hard to keep track of. It was still orders of magnitude better than doing it by hand, but this is just that next step function in terms of productivity gain in my mind.

Um, so that's, that's why I was super excited about when I saw it.

Mallory Mejias: And as we've mentioned with canvas, these inline edits are really exciting. So with Claude's artifacts, you can in the chat say, ah, this bullet is not exactly what I want. Can you change it ever so slightly, but it's kind of time consuming. And now seeing what this new option is, which is just highlight, add a quick note of change, thus just this bullet point, um, that's exciting.

Now, I still think I like the output for me personally that Claude creates, but this gives me a lot more flexibility. Um, and then I can do that with infusing my own thought, um, into what I'm writing. So, uh, that's very appealing. Have you tried this out with, uh, code creation at all?

Amith Nagarajan: Not with code yet. I think that'll be a natural next step for it. Um, a lot of times when I'm doing coding, I'm working within the, the visual studio code or VS code environment and working with co pilot in there. So there's a co pilot built into the code editor. Um, sometimes if I'm doing coding, I'll go to chat GPT or sometimes other tools and ask it to create new code based on some general ideas.

And then once you're more into like the refinement process, you're in the code editor and Copilot in there is really, really good at helping you do kind of the incremental stuff. Uh, so I do think it has applicability for coding. Um, but at least in my workflow and what I tend to see other people do that are doing similar work is to do your kind of your big chunks of new work in a Chachi PTA type environment.

And then you pull it over into a code editor and then you're doing more incremental work there. Uh, interestingly, there's been a lot of talk recently about new tools for software developers. Uh, some that are fully automated, that are things that are capable of like doing full end-to-end software development, and some that are in ides, like Visual Studio.

Uh, there's another one called rept, which is very po uh, popular, that has really great, uh, copilot style. Assistance. There's a new one called cursor, which is actually a fork of V. S. Code that has enhanced a I capabilities like built into the environment. And I think you're gonna see more and more of that with applications is that the app that you use is gonna have more and more kind of a I native workflow built into it.

We've been talking about that for a long time. How if you're in Canva, it makes sense to use Canva for what it's good at and have a I tools built in and that particular product. They've done a lot of that. That's true for the Adobe suite. That's true for obviously Microsoft with Copilot. You're going to see these tools essentially become part of the workflow.

I do think the leading edge tools like, uh, like, uh, Cloud and ChatGPT, uh, do have a place. And they're going to have more power probably than some of the embedded tools that you'll find, but it's going to be interesting to see, like, will you use chat GPT at all in three years? Or will these capabilities just be woven into other productivity tools?

You know, as chat GPT become the super app that can do spreadsheets and presentations and graphic design and word documents, um, or is it, you know, more of a companion tool? It's going to be really interesting to see how. How that shifts over time, um, for associations, one thing that you should be thinking about is how does this affect member experience?

So if you think about the way you engage with your audience, how are they engaging with your content? How are they using your content? Um, a lot of times associations put out a bunch of different products. They'll publish journal articles. They'll have in some cases like books. Um, they'll have obviously meetings and webinars, all of these different essentially knowledge products of various kinds.

Um, what do people actually do with those resources when they join your association or just buy individual resources? What's the point of what they're doing? Do we really even know that as associations? Have we done the user research to find out what happens After one of our customers or members actually access is one of our information resources.

How are they using it in their day to day job? And the reason that's important to understand is what can we do to make our resources more valuable? Do we want our resources to be something that could essentially have like a? A canvas equivalent where, you know, users can directly interact through your association to make changes to make it work better for them, right?

Using a I tools directly as part of their engagement experience of the association to still be in your context with all the value you provide. But be able to adjust things to hyper personalize, uh, that content and that engagement experience for their needs. And I think that a tool like Canvas should be a thought experiment for associations to look at their own resources, their own software offerings.

And by the way, everything's a software offering, you know, whether it's a downloadable PDF, or a blog you read, or a webinar, they're effectively all In this modality, how can we think about our engagement experience to shift from a chat GPT, you know, classic style of engagement to canvas, right? Those kinds of shifts reframe the expectations of the broader public.

And that includes your members. So if you don't think about how to engage people in ways that are contemporary and aligned with what they're getting in consumer grade tools like this, you're probably going to end up in a, in a bad situation at some point. So it's an opportunity to draw inspiration from it.

Mallory Mejias: One AI knowledge assistant that we've talked about on the pod before is Betty, which is, um, in our same family of companies. Amith, have you thought about adding some sort of, uh, canvas like feature within a knowledge assistant like that? Potentially?

Amith Nagarajan: I think Betty having a canvas type feature would make a ton of sense. I think that's true also for Skip. Skip is our AI data scientist that generates reports. Um, skip already kind of works in that modality a bit. Um, but this idea of a canvas or essentially like a workspace, I think makes sense for those kinds of applications.

Pretty much most everything, if you think about it, because, you know, we've had this, I think the chat modality of. Conversational interaction is awesome, but ultimately, you know, whether it's an artifact or a canvas or a document, whatever you want to call it, You're usually creating something like that, right?

You're creating some kind of durable result of the conversation. And so what we've had so far is just this long running chat and you've had to like copy and paste, like out of the chat, drop it into a word document or a Google doc, and then do the final work there. And they're simply streamlining it saying, What chat GPT folks, their product management team is probably saying is like, what's the use case?

Like, how are people actually using this? How can we further streamline and make it more efficient, add more power to it? Um, and so that's, that's where I see this happening in general. So one thing I would suggest, um, is, and I'm sure this is on the roadmaps to these products, is multi user collaboration along with the AI.

So if you say, okay, You know, we can share, like if you had that chat thread that you just had in chat GPT, you can share it with me and I can look at it, but I can't interact with it. So it'd be really cool if you had like a Google Docs style multi user collaboration where chat GPT was in there and you know, you could tag chat GPT to make a change or we could tag another person to review it.

Um, that really would blend the workflow that a lot of people have gotten used to with Google Docs and Microsoft 365. And of course, Copilot in the Microsoft realm is going to do exactly that and is already doing exactly that in many respects. So, I know that's a little bit of a preview of our next topic, which we can get to in a minute.

But, uh, I find these tools really fascinating because it's going to fundamentally change the workflow.

Mallory Mejias: Yep, we will be talking about exactly that. Copilot pages in the next topic. But one more question here, Meath. I wanted to spend a little bit of time talking about content transformation because when I demoed and shared my screen we saw there were these built in shortcuts essentially to help you transform your content from long to short or adjust it based on reading level.

I would say this is In terms of having one piece of content that you can reuse over and over again, but I kind of wanted to have a little discussion on the, I, the strategy behind that and the idea of quality over quantity using the sidecar sync podcast. For example, we've talked about how we use the transcript from this podcast to create blogs.

Theoretically, we could create. many blogs, I'd say three to six. We could honestly probably create more than that. If we broke down the topics into subtopics, we could create 10 social posts, um, a six step email sequence, 10 social clips, a whole course based on every podcast episode, we could do all those things in theory, but perhaps we shouldn't.

I don't know. So I just wanted to talk to you briefly about kind of now that content transformation is really so easy. It's at our fingertips, how you would recommend approaching that.

Amith Nagarajan: Well, you know, when I think about it, I look at and say, Are we trying? What's our objective? Are we trying to reach a broader audience? Are we trying to engage our current audience more deeply? Maybe both. Maybe there's other goals. So let's talk about each of those. Dynamics a little bit separately. So let's say we want to reach a broader audience.

So for example, let's say that our association is an association of electrical engineers. And so in our domain, we have a lot of very technical content about electrical engineering, and that's our core audience. Is this, you know, center of, uh, of the Venn diagram, right? It's this. Electrical engineering group.

And so our content is written by electrical engineers for electrical engineers. But let's say that we need to collaborate with some other, uh, domains. Let's say there's some scientists out there we need to work with. Um, and they're not electrical engineers. So their language and knowledge is a little bit different.

Um, say they're physicists. and we have physicists and electrical engineers now working together. Or maybe there are mechanical engineers who we need to collaborate with. Uh, so there's still obviously all very technical people, highly educated people. So it's not about a reading level, but it's more about domain knowledge.

So can we take a journal article written for electrical engineers and then make it more comprehensible to say a mechanical engineer who's at the same level of Intelligence, education, all that kind of stuff, but comes from a different domain. That can be interesting. Um, so that's a form of content transformation where we're trying to reach adjacent, uh, professions or adjacent verticals that I think AI could tremendously help with.

Because if you have an AI that's an expert in your domain, and let's say there's another AI In this example, let's say there was an AI for the Mechanical Engineers Association and an AI for the Electrical Engineers Association. And those respective AIs were trained on the corpus of content from each association, so they were domain experts.

And they were able to collaborate to say, hey, let's take the Electrical Engineering article and the Mechanical Engineering article. Agent, if you will, and the electrical engineering agent could then collaborate to say, Hey, let's make this really relevant for the mechanical engineers. So the mechanical engineering AI would say, Okay, this is what I think might make sense.

And the electrical engineering AI would say, Is this still correct? In the context of what we're trying to communicate, and then the mechanical engineering AI would look at it another pass and say, Yeah, does this will this make sense to my typical audience? Right? So there's like a multi agent collaboration going on there to do a domain across domain kind of knowledge transfer.

Another example might be, let's say that we're a dental association, and so our primary audience are dentists. But people that are super, super important to, um, you know, basically oral health generally are dental hygienists. folks are actually doing a lot more spending a lot more time with a patient and dental office than the dentist.

Um, so let's say we have some content from the dental association that is dentist centric, but we want to make it available to the hygienists, which have their own association and their own content. Same kind of idea, right? In that case, um, maybe it's a different tier of the profession, right? Or medical assistance relative to physicians and so on.

So I think there's a ton of opportunity here to This could open up content for more use. It could improve the quality of care in the case of healthcare world, uh, improve the quality of engineering in the engineering world. So it's, it's really exciting. Um, I focused there initially in describing this because the most obvious example of content transformation is one language to another.

So saying English to Spanish to French to German to Dutch to whatever. Uh, and that's of course a wonderful capability AI gives us. Uh, there a domain specific language model Um, it's also important because you want, you know, your mechanical engineering text to make sense in French. Uh, and a lot of generic LLMs, like whether it's open source or chat GPT, may or may not get that right.

But if you have a French mechanical engineering LLM and an English mechanical engineering LLM, or preferably one trained on both languages, you're way more likely to get a great transformation outcome there. So those are just some use cases that I think are exciting. Um, and to your earlier point about transformation in terms of like, Taking bits and pieces of a piece of content and repurposing them.

I totally agree. That's a really exciting area and you have to think about your strategy and say, okay, why would we break it up into so many chunks? Is that to try to personalize the way we reach out to people and say, Hey, Mallory, we really know that you're super interested in, let's say, Email strategy.

And so we bring you more content that's related to that particular topic. And this guy, me, if he didn't like email, he's more interested in social media, give him more of that, right? So like a classical personalization strategy, that's certainly one way to leverage content transformation is to bring out those pieces.

Um, anytime we can save people time. I think that's interesting, right? Because everyone's busy and so like the shorter we can make something and then let them choose to go deeper into it, the better. So, summarization of course is another now classical example of what language models are really good at.

So hopefully that helps a little bit.

Mallory Mejias: sure. So before going out there and transforming all of your content into various other formats and modalities, ask yourself why it sounds like that's the essential question. Why are you doing it? Are you trying to broaden your audience? Are you trying to personalize the way you reach your members and then go from there?

Amith Nagarajan: Yeah. And, you know, another aspect of transformation it's worth noting, just to link this back to a recent episode that we did an unstructured data is transformation from unstructured to structured insights. And we talked a lot about that on that pod episode, which was a very popular episode. And I think the utility in that is quite high where you can say, okay, for my own content or for other people's content, can I answer certain structured questions about that content on a consistent basis using techniques like we described in that other episode, which will link to in the show notes, um, where we talked about how can language models actually be.

Go through, read the content or watch the content and then answer a set of structured questions. Um, that might be very instructive in terms of a lot of different applications you do downstream from there. So the examples we showed we talked about in that other pod, for example, were Take a corpus of research papers and answer a set of questions like generating metadata essentially from those papers.

That's a form of transformation worth noting as well. And you can do the inverse as well, where you can say we want to generate content based on these topics, use our existing corpus of content and generate a new article based on these five or ten topics.

Mallory Mejias: Amith, editing break. Do you have a hard stop at the hour?

Amith Nagarajan: Let me see. I am good for another 15 minutes past the hour.

Mallory Mejias: Okay, I don't think we'll go much over that, but okay.

Moving on to topic two, Microsoft Co-pilots Wave two, which introduces several significant updates and new features to enhance AI powered productivity across Microsoft's suite of applications. One of the new rollouts is exactly what Amith just explained basically, and that is co-pilot pages, which is a new collaborative canvas designed for multiplayer AI interactions.

It allows users to pull insights from work data into an editable document, to share and collaborate on AI generated content with your colleagues, and to iterate with Copilot like a partner, adding more content from your data, your files, and the web. So it's essentially Chatubty Canvas, but built into, into your Microsoft suite.

With Excel, we're seeing more support for formulas, data visualization, and conditional formatting, as well as Copilot in Excel with Python, which enables advanced analysis using natural language. In PowerPoint, we're seeing a new narrative build, A new narrative builder feature for creating first drafts of presentations from prompts.

I actually went into PowerPoint before this and realized we have this available to us now. So I'm going to be testing that out. And within PowerPoint, you also have a brand manager to ensure CoPilot works within your brand templates. In Outlook, we're going to see a feature called Prioritize My Inbox to help you manage emails more efficiently.

It can do things like recognize who your manager is and who your key contacts are and it then ranks your highest priority urgent emails based on that. Within Teams, we're seeing improved meeting summaries, incorporating both audio transcripts and the chat content. So normally in meetings, I feel like within the chat interface, we'll have things like questions pop up or little notes.

People might not want to unmute, so they'll add it into the chat. So now with these summaries, it's actually going to reference all of the chat information, as well as the transcript from the call, which is really exciting. In OneDrive, we're going to see better ability to find files, Generate summaries on your files and compare documents without opening them.

So the example that they gave in the video was when you have two files that have very similar names and I am guilty of this. You can actually just ask Copilot to compare them, to highlight the differences in the two documents without opening them. That sounds incredible to me. And then perhaps, maybe the most exciting, um, is Copilot Agents.

So Microsoft's introducing Agents. Agents, which are AI assistants that can perform specific tasks with varying levels of autonomy. These range from simple prompt and response agents to autonomous agents that can actually take action. They can be built using the new agent builder powered by Copilot Studio and they're accessible in Microsoft 365 applications using the at mention.

They can also be in your team's chats and can take action if prompted, which I think is exciting. Copilot now uses OpenAI's GPT 4 0 large language model. Responses are more than two times faster on average, and Microsoft says response satisfaction has improved by nearly three times. So, Amith, I watched this video with the initial rollout, but I watched it again yesterday to prep for this podcast and It was really exciting, but it, it makes me think about the initial rollout of Microsoft co pilot when we watched the video and we said, this will change everything and admittedly you and I, I would say are not the biggest co pilot users.

So I'm curious to hear your take on wave two.

Amith Nagarajan: I'm excited about all of the things you just mentioned. You know, I think the outlook prioritize my inbox piece, I think is amazing. You know, I had a, a former business partner of mine in a, in a different company long ago, once had his own, this is years ago, he had this prioritization strategy, which is pre AI, where he basically didn't respond to anybody's emails at all.

Until they responded, until they emailed him again saying, Hey, what's up with this? He's like, Oh, if it's important, they'll email me twice.

Mallory Mejias: Wow.

Amith Nagarajan: That may not be the strategy that outlooks AI does. I hope not, but, uh, it, it seemed to work for him at least for a while until people got really upset. Um, But even I think everyone suffers from email delusion.

So having help with that will be great. The co pilot pages concept you talked about is exactly, you know, what I was describing earlier, where you can have a multi user collaborative environment. And then with the agent capability, we've talked a lot about agents on this pod and the rest of our content, uh, being able to take action based upon what's going on.

So imagine. A collaborative space where you're working on a new blog post and you have a couple of people that are collaborating on it along with copilot itself. You get the blog to where you want it and then you say at and then you tag an agent, which is your publisher agent. You say, go ahead and publish this to my help spot, C.

1. S. Um, and then that agent, we've trained it to also automatically break it up into five or 10 social posts. and do all those other things downstream. So that's a good example of how you can add your own custom agent using the agents toolkit that you mentioned, um, and connected to other applications.

That would be a non trivial exercise to set up all that. But it's possible to do now within the workflow described here. Um, something that's probably a little bit lower on a lot of people's radar but super valuable is this idea of code generation in Excel with Python. So, uh, Python for those that aren't familiar is a very popular programming language that is particularly good for data analysis.

A lot of people that are in the machine learning and AI space use Python as their primary coding language to both train models and do other kinds of data analysis. And the ability for Excel. And copilot and excel to interact with the user and then just generate programs that can do stuff with the data and excel potentially could be really a game changer for a lot of people.

So you imagine someone in your finance department that's working on a financial forecast and they want to do something beyond their capability. You just ask copilot. It might generate some formulas, it might use pivot tables, or it might actually go and write code in Python. Uh, so that, that's a very powerful thing to explore.

Um, PowerPoint scenario that, um, I think has a lot of potential because, you know, and I've actually, this is the part of copilot I've probably used the most over the last, whatever it's been six, nine months since we've had it, um, where I'll take a word document. After I've published something and I'll say, Hey, PowerPoint, here's the word doc generate a set of slides from this document, and I found that to be very effective, even in the current version of copilot.

Um, so those are some of the things that I'm probably most excited about. Um, To me, um, I think the better model is going to be the main reason to start using it. So I've found that Copilot's original release with the, I think it was GPT 4 Turbo that they were using, um, was underwhelming, um, because we'd moved on in ChatGPT and in Anthropx Clawed to much more powerful models and the version they had in Copilot seemed to be underpowered.

So it was kind of like, you get used to, you know, Flying on an Airbus A380 at 570 miles per hour and then you have to go back to flying around in a Cessna at a 250 miles an hour or whatever. So, you know, you're not going to want to do that. That experience doesn't feel right. So, if the model level within Copilot is at parity with what you get in CHAT GPT, I think it will get a lot more natural usage.

The other part of adoption though is habit. Right? So we're not in the habit of using Copilot in Word and in Excel and in PowerPoint. We are more in the habit of using an extra tool like an Anthropic or a ChatGPT. So I think if we kind of push ourselves to try to use Copilot more, it'll help. Because having the tool woven into the primary, you know, workflow tool that we're using for day to day work, um, I think should be able to create efficiencies for us.

So I'm excited about it. I think Microsoft in recent years has done a tremendous job with their marketing, that original co pilot launch they did last year was amazing. Um, the product itself, I think, you know, holds the promise of living up to that video eventually, but obviously it's not there yet.

Mallory Mejias: I have full faith that it will eventually be just as it was in that video. We need to get them to make a video for the Learning Hub because that was truly impressive to watch. Um, and I'm going to put it out there as a, as an accountability system. As all of these new features roll out, I want to create a Microsoft Copilot course for the AI Learning Hub.

We get tons of feedback and questions all the time, whether we have that content. And so I'm making the commitment right now on the side cursing podcast that we will be doing that. Within the next few months. Amith, my question to you is what, assuming this works exactly as predicted, what happens to all of our beloved tools like beautiful.

ai, Midjourney, even ChachiBT, um, do you think we're entering an era where we need to be prepared to potentially say goodbye to these tools? Yeah.

Amith Nagarajan: waves of software innovation, where things start off as standalone products and apps, and then they really ultimately become features. In some bigger platform that's happened for years and years and years and you know It's it's good to have an explosion of capabilities as all these little apps but then ultimately a lot of them, you know kind of get sucked into the capabilities of what a Mainstream broader tool is going to use so, you know something like beautiful that ai.

Um, I don't know, you know You have a lot of great Tools that can do a PowerPoint style presentations, right? And a lot of people have tried to kill PowerPoint over a lot of years and had a hard time doing it because PowerPoint has user base. It's kind of clunky in some ways, but it works. Um, and if PowerPoint is AI powered, why would you go to beautiful?

I used to use beautiful. And then when I got copilot and PowerPoint, I stopped using beautiful. I mean, for me, like I was a little bit less interested in some of the design features they had there. Like I'm. A very basic PowerPoint user in terms of the design side. Uh, but for me, PowerPoint, even with the first version of co pilot was plenty of horsepower to do what I wanted.

Um, so I don't know. I think it is natural though, to see some of these tools go away. I think the first casualties will be some of these AI note takers that are out there like meet geek and products like that. Not necessarily that one specifically, but you know, you really don't need them like in zoom and in Microsoft teams and in Slack, you have noted note takers.

Taking summarization, action items, all that stuff built into those apps. Yeah. People keep using read that AI meet geek, uh, all these other tools, uh, which are also potential cybersecurity issues because they're an extra tool that's kind of jumped into your most sensitive zoom meetings. So I think some of those things are very obviously just.

Features and go away. Like, uh, actually a good example early in the chat GPT explosion was, uh, talk to my PDF type apps, there were a whole bunch of apps that were like, Hey, talk to a PDF because chat GPT to upload a document. Well, gee, really? You think that's going to take too long for chat GPT to add?

It's ridiculous. So of course that had utility for five minutes and then was wiped out. So there'll be some of that. And I think that's ultimately good for consumers because that type of competition is going to drive more value creation in the, in the core tool set. Now, as far as will chat GPT itself go away, I don't, I don't think that that's likely.

Um, I think that they specifically have enough of a user base that they can add feature set to chat GPT that might make you ask the question, does Google docs go away? Does Microsoft 365 go away? And those capabilities become part of a suite of tools that chat GPT makes available. I could see that happening too, you know, cause think about what's happened with artifacts in a, in Claude and canvas that you just demonstrated.

In chat GPT, you add just a handful of more features there for editing documents. And then you make the UI capable of showing you all the documents you've created outside of, so you take all the different like canvases that you've created in chats, but flip that around and say, Hey, there's a document browser and chat GPT, where you can see all the, all the canvases or documents that you've created.

Um, and then being able to go back that way and to be able to share them, collaborate in there. It's not all that different to Copilot Pages or Word documents or Google Docs. So, you know, Chat GPT has enough momentum where they stand a chance of being able to drive tooling like that. And they have one really significant benefit, which is that no one expects anything from them.

So they can add the simplest, easiest features. Whereas Docs to a large extent now, are the incumbents, and they have a ridiculous number of features. So weaving in AI, you know, is kind of hard actually when you have that much, um, you know, what I'd call kind of legacy feature set that, you know, 5 percent of the users use, like, you know, most of those features.

There's, you know, a million features in Word or whatever, and most people use 10 of them. And then the other, you know, other, uh, features that are out there are used by a tiny fraction of people.

Mallory Mejias: I'm going to veer a little bit because we have your expertise as an entrepreneur, Amith, but you said history has kind of repeated itself. We've seen this before where companies built around this one feature go away because that essentially becomes a feature in a bigger platform or application. So why?

Why is it then that we saw kind of the appearance of like meeting note takers, if we're going to use that as the example, in the first place, if we knew kind of in a few years that would be rendered useless?

Amith Nagarajan: Well, entrepreneurs are always going to chase what the next opportunity is, right? That's what entrepreneurs do. We're wired that way. And so we look at it and say, Hey, um, where is a missing piece of value creation or where is there an efficiency or, you know, where can we use a new technology or a new platform to create value and then monetize it essentially.

And so the idea of a meeting note taker is amazing. A. I'm eating note takers that have been out there for a little while. Um, have built considerable businesses. Um, so there's an opportunity to rapidly build a tool like that and then sell it to someone who will then weave it into, it'll turn into a feature in their product.

So that's what oftentimes the playbook is to move quickly enough where you as the innovator, as an entrepreneur, Can rapidly build something. And then the idea is, is that it's likely that one of the platform players will say, Hey, let me just buy that thing and make it a feature. That's one possible playbook.

Uh, the other is to basically have this ego big enough where you think that somehow you can buck that trend and become the next platform. Uh, and that occasionally does happen. It's a very low probability play. You know, it's very unlikely for anyone to win doing that. Um, And then there's other strategies that are out there, like what, you know, a large part of my career has been focused on is verticalization, where you take a capability, um, and then you make it hyper specific to a narrow, uh, go to market or a vertical like the association market or the nonprofit market or other companies that I've been involved in have been in other verticals, whether it's, you know, home healthcare or construction or a number of other verticals where I've been involved as a founder or investor.

And so. Um, I like these hyper specific narrow markets because that's where the domain expertise, the relationships, the market knowledge is actually a really significant value add. I mean, it's why products like AMS's exist. As much as people don't like them, um, they provide a level of value creation above a generic CRM.

that associations largely need. Um, and so that's a reason why a lot of vertical products tend to have an interesting opportunity, but coming back to the horizontal plays, um, like a meeting note taker, they tend to be very short term opportunities and either catch that opportunity early enough, get enough market share and exit, or you get crushed.

And I think most people who do that kind of stuff know that, but they're, they're playing the odds.

Mallory Mejias: That's very helpful. I want to ask one more question on Microsoft Copilot Wave 2 and it is around agents. So in the video that they rolled out, it seems really intuitive to build your own agent. They like walk you through the process really quickly. Uh, anyone essentially will be able to build an agent this month, I think, is when they're rolling this out in beta.

And on one hand, that's amazing. And on the other hand, for me personally, I'm thinking, well, What agent am I going to build? I could build any agent I want. Where do I start? And so I know you often talk about going to pain points I can think of a few that I have but I just wanted to to give you an opportunity to speak about now that Potentially anyone can make an agent this month how you would recommend going about that

Amith Nagarajan: Well, I think C seeking out pain points, um, which another way to put that is looking for inefficiencies, looking for things you do repetitively is a great opportunity for agents. And so the email use case, I think it's a great one. Most associations get tons and tons of emails that are basically the same thing over and over again.

And whether or not you have an FAQ in your website, Uh, and most people do, they still get, you know, 50 percent of their email volume or 80 percent of their email volume could be answered by looking at the FAQ or looking at the top 10 most commonly cited documents. So there's definitely an opportunity for a knowledge agent.

like a Betty bot or or others like that to be able to be wired in to an email agent that can respond. In fact, the member junction team has exactly that in the works and a digital. Now we're gonna be announcing service agent capabilities where Betty will be able to wire into your email and automatically respond for you.

Um, and that's similar to what you can build on your own using something like co pilot agents. The main difference is whether you want to Have an enterprise scale type of approach where you have all your entire corpus of content grounded truth in terms of responses. That's important for a knowledge agent that's going to answer questions, let's say, within the domain of your association.

So back to our engineering examples. If someone is just asking, you know, what, what's the location for our next annual meeting, basically any AI can answer that from a couple of documents. That's easy. But if they're asking a question about electrical engineering or about mechanical engineering, you do not want to get that wrong.

You don't want to use a consumer grade tool for that kind of a question. So that's where it's more of an integration play where you, you pull in like, uh, An enterprise grade knowledge agent, like like a Betty or something along those lines, and then wired into your email infrastructure. Um, I think email is a great place to start because you can improve the speed of your responses and give people better, more comprehensive responses faster, and you can lower your workload.

So, you know, you went on both sides of it. Better member service and lower internal cost.

Mallory Mejias: Awesome. That's very helpful. I am excited to see wave two of co pilot to make a course on it and hopefully for it to live up to that initial video that they showed. Amith.

Amith Nagarajan: exciting.

Mallory Mejias: Yep. Thanks so much for joining today. Everyone, our viewers on YouTube, our listeners on all major podcasting platforms, we will see you next week for episode 52.