224 | One AI to Rule Them All? I Tested Gemini vs. ChatGPT Across Real Business Use Cases

0:00Hello, and welcome to the Leveraging AI podcast, the podcast that shares practical and ethical ways to leverage to improve efficiency, grow your business, and advance your career. This is Isar Maitis, your host, and we've got a very interesting episode for you today. The reason it's interesting is it is based on my own personal experience that is tied to a workshop I have delivered two weeks ago. So two weeks ago, I was invited to provide a several-hour workshop at a very large, really known tech company in San Francisco. The company that invited me uses Google as their office infrastructure and hence the request was to focus most of the workshop on Gemini as well as other Google AI tools.

0:39And that's what I did. But to do that, I had to take a lot of the workshops that I've been teaching to many other companies and customize it and test it on the Gemini environment, which gives me a very unique insight to how it is compared to other tools, especially ChatGPT. Now, I've learned a lot in the process, and I thought it's going to be highly valuable to you too, to learn from my experience from what I've learned from trying to convert use cases from Claude and Chachupiti, mostly Chachupiti, into the Gemini universe. And so this episode is not going to be structured in a very specific flow because these are unique specific use cases.

1:15We're just going to go use case by use case and compare what the results were and what were the pros and cons of each and every one of these platforms. There is a winner, which we're going to get to in the end, but let's get started with the comparison. In the next few years, AI technology will change our world dramatically. Whether you are a business executive trying to catapult your business forward, or just somebody who refuses to be left behind and want to advance your career, this is the show for you. I'm your host, Isar Maitis, a serial entrepreneur and an AI enthusiast. You'll hear invaluable practical tips from innovative business leaders, AI practitioners, and some of the brightest AI minds in our world today on how you can leverage AI in ethical ways to advance your career and grow your business.

2:10The first thing that I want to talk about is a concept that is critical for the success of anything you do with AI, and then that is the concept of context. These tools lack context, right? They don't know you, they don't know your company, they don't know your industry, they don't know your unique expertise and because of that a lot of the answers you're going to get are generic so the more the tools know about you the more information you provide them in order to learn about your universe the more specific the more accurate the more relevant answers that you're going to get that is true for all these tools which means that if you want to get accurate answers you need to write very long prompts every single time unless you can teach the model stuff about you that it will remember for the long run and that is a very big difference right now between Chachupiti and all the other models, but in this particular case, Gemini as well.

2:57So let's look at what I'm referring to. So what I did here is I basically asked Chachupiti, what do you know about me? A very simple and straightforward question. You can do that yourself to test it yourself. You can try it with different models, but let's see what Chachupiti knows about me. And this is just a question that's, there's nothing else in this prompt. And it says, here's a structured overview of what I know about you based on our past conversations and context you've shared. background. Former F-16 pilot in the Israeli Air Force, later an instructor in the Air Force Academy. Transitioned to business and technology with 15 plus years as a CEO.

3:29Experiencing startups, people post, and large corporations. Last minute travel, hotel beds group. Moved to the U.S. about 20 years ago. Professional journey. Built and scaled multiple companies, including taking one public and selling another. Founded Multiply in 2023, focused on AI education and consulting, provides and so on and so forth. Then it has content education, which are the things that I'm focusing on right now. It talks about this podcast, the Leveraging AI podcast, about our newsletter, about the Hangouts we do every single Friday, and about my collaboration with the Crummer School of Business, which is the business school I did my executive MBA at, and now I've been collaborating with them for the past year and a half in delivering AI education to their audience.

4:07And then our other initiative that I'm involved in, Market and Audience, where it is, who am I'm catering to, personal notes about myself, about my family, what I'm passionate about, what my family does, and so on, and some key takeaways. All of that, it knows about me, which means when I'm asking it anything, it uses that reference information in order to provide more relevant answers to me. I've done the same exact thing with Gemini and asked Gemini what it knows about me. So same question, what do you know about me? And the answer is, based on our current interaction, here's what I know, just as important, what I do not.

4:38What I know, our current conversation, And then it says what we talked about in that particular conversation. Time and location based on the information provided to my system. I know it is Thursday, August 7th at 10.49 a.m. Technical information, very little. And what it doesn't know, it doesn't know my name. It doesn't know my age. It doesn't know where I live. It doesn't know my email address. It doesn't know my browsing history. It's a lot of other stuff it doesn't know. So basically, it doesn't know anything about me, which is good from maybe a security perspective. It's bad from a context perspective to any conversations I'm going to have.

5:10So I actually take the time for a very long time now, investing in teaching ChatGPT about me, about my business, about my tone, about my customers, about the proposals that I write, about the services that I deliver. It knows a lot more than it actually shared in that one summary. And because of that, it can provide me much better information. The other thing that I use for context in ChatGPT that does not exist in the Gemini universe is I use projects. projects are on the left side right under the custom gpt's segment just above your chats and in there you can create these folders and in these folders you will have different conversations but each conversation can have its own unique context universe so you can see in this particular case this is a sample client project it's not a real client but i have one for each and every one of my clients in there you will see that i have different kinds of information i have proposals for that particular client that I've sent this client.

6:01I have different files that I've created. So the culture of the company based on my experience and based on their website, what proposals I've sent them so you can see what that is, the consulting agreement I have with them, the people in the company that I engage with and their profiles from LinkedIn, the workshop proposal that I sent them, the about us PDF from their website. So more context about that particular client. So then when I have conversations in this folder, it has a lot more context about this particular topic, which again is not something you can do in Gemini. So that's on context, which is definitely a huge win on the chat GPT side.

6:34So let's move on to the next use case. The next use case is using custom GPTs versus using gems. So the concept is very similar. I'll explain in one sentence what it is. So custom GPTs and gems allow you to write instructions that you want to repeat time and time again. So every time you have a repetitive thing you want to do with AI, whether it's writing a proposal, analyzing a specific document, comparing two Excel files, creating content, et cetera, et cetera. Every one of those can become either a custom GPT on the ChatGPT universe or gems on the Gemana universe. I recorded an entire episode about it.

7:08And if you want to dive into the topic of custom GPTs, you can go back to episode 175. It is called Stop Wasting Time, Automate Repetitive Tasks with Custom GPTs. So you can go back and listen more about that topic. But there is a similar concept that is more or less the same in the Gemini universe called GEMS, where you can provide information and so on. So the way I create and the way I teach how to create both custom GPTs and GEMS is not by writing the instructions on your own, but rather by having a conversation with ChatGPT or in Gemini and then asking ChatGPT at the end of the conversation, once you explain to it everything that you want, once you ask it questions, once you work inside the conversation on the actual output and you achieve the correct output, then I would go and say, I would like you to now turn everything that we've done, including the final successful outcome, so ignore the stuff that didn't work, into highly detailed instructions for a custom GPT or a gem because it creates better instructions than I can, especially after we've done the entire exercise and we're getting to the outcome that we want.

8:05And I've done this in chat GPT multiple times and then I tried it in Gemini and it doesn't work as good. So let's look at an example from each and every one of those. So in this particular case, this is from a presentation I've done for the Israeli parliament. And I've started the prompt with, you are an expert data researcher in the Israeli parliament, the Knesset. We are planning to use a custom GPT to help us in the research and summarization of data. We would like to build a team of separate personalities that will participate in this process, et cetera, et cetera, et cetera. I'm writing in this prompt.

8:34And then we're going back and forth and it gives me suggestions of what these people can be and what they will do and what are going to be the responsibilities and the roles of each and every one of them and what's going to be the research process and what's going to be the focus on different things. And then I ask it for a list of responsibilities for each and every one of these people. So I went ahead and did that. So now I have a long list of responsibilities. And then in the end, after going through the entire process, I said, please create a detailed instructions for a custom GPT that will create all these personas.

8:58The GPT should receive an input that would be a topic for the Knesset team would like to research. The team should first discuss the topic, debate, and agree on the way forward, et cetera, et cetera, et cetera. And it wrote these amazing instructions. So you can see there's a system description. This is the first thing that it created. Then it created a team overview. It gave names to each and every one of the people. What's their role? What's their style? So they can have different personalities that will support the team. Then a step-by-step workflow. Step one, topic intake and group discussion.

9:26What are the actions that they should follow? Step two, data collection. Step three, domain analysis. What each people, which, and it even has a breakdown of who's leading what aspect of the process based on the personas and the roles and the actions that they need to take and the responses that they need to provide. Step four, political and public lens. Step five, quantitative analysis. Step six, synthesis and drafting. Step seven, risk and ethic review. And then who is the lead on that and the actions for all of that. Step eight, final review. And then there's a communication format on how they should communicate with one another.

9:58And then what is the output structure of the report that they should generate and additional capabilities. So this is an incredible set of instructions that, again, there is no way I could have written this as accurate and as detailed. And then I did a similar process in Gemini, asking it to create instructions for a specific business process, not exactly the same. And when I did that, it actually gave me instructions that were okay, they weren't horrible, but they definitely weren't great. And it was weird because it was kind of like relating it in third person. And I'll explain what I mean. So my prompt was, I would like you to create instructions for a gem that I can run inside of Google Slides that will ask the user this set of questions and topic, blah, blah, blah, blah, blah.

10:33And then it created a set of instructions that should be able to work, but it invented things that just don't exist. So as an example, it said this gem will open in the sidebar in Google Slides. It will guide the user through a series of five steps, asking questions and so on and so forth. But the reality is it cannot open on the sidebar. So I wrote back, I said, I don't believe the sidebar can have a next button. I believe it is a simple chat interface. Am I wrong? But then it said, you're right to question the specifics as the interface isn't free for in Canvas. However, it's more powerful than a simple interface and you can absolutely have a next button, which by the way is not true.

11:09At least I wasn't able to see it true, but it gave me instructions that just don't work. So instead of just giving me instructions for a gem, it's referring to what needs to be the instructions in the gem, which is not horrible. And in many cases it actually worked okay, but it's definitely not as good as the solution that I got from ChatGPT. Then I tested something else. I tested a very complex multi-step custom GPT that I've created a while back for a client. And I was trying to recreate it with the same exact instructions on Gemini. So let's look at that. So before we dive into this, let's talk about what this multi-step process does.

11:43And it's actually going to blow your mind just by itself, because what it does is it takes a inventory file that is completely fake. I made it up. All the numbers are made up, but it's an inventory file from multiple sources showing inventory of different SKUs in different locations for the company, and it is using it to create a sales brief for the salespeople to explain to them what inventory we have, what inventory we're out of, in which regions of the world, and so on. So this is something that many companies struggle with, right? So the ability to take information and compare the inventory from yesterday to inventory from today, the inventory from last week to the inventory from this week, and based on that, update sales information that then can be sent to salespeople.

12:25This is a process that many companies do completely manually today. And I've developed a custom GPT that does the entire process, but it's a five-step process. So all I have to do right now is upload the old inventory file, the new inventory file, and the old sales file to the custom GPT. And then it follows all the different steps. It does step one, step two, step three, step four, compares different things, where it is, what are the sizes, what are the SKUs, if it's the same package type, and so on and so forth to verify all that information. and then when it's done with that step, it goes to step three and then it removes all the roles that do not have the relevant roles in the other file anymore.

13:00So basically we have it in the sales files but they don't exist in the inventory file anymore. So it compares three different files back and forth. It updates all the quantity and the numbers and the SKUs and everything to provide me a new updated file and then it provides it to me as a CSV file that I can download to either upload to whatever CRM system that I'm running or to email to my team. Again, the CSV file can then be converted to anything that I want. So this is an extremely powerful multi-step process that I have tested consistently multiple times and provides accurate and amazing results every time that I run it.

13:35You can imagine similar processes across multiple aspects of businesses where you have multiple data sources that you need to align, compare, and provide some kind of an output based on all of them. And it works flawlessly inside of ChatGPT. However, when I try to run it on Gemini with the same exact instructions, it literally gets stuck. It doesn't do anything. It says, of course, I will process the files according to the steps provided. Please provide a moment while I perform the data comparison, filtering, merging, and updating. I'll provide you with a final updated file for download shortly.

14:06That sounds very promising, only nothing happens. And so I tried several different ways. I changed the instructions a little bit. Again, the instructions were copied and pasted, and yet it does not work. The inputs are the same exact files that I gave ChatGPT. So in this particular case, again, a huge win for ChatGPT compared to Gemini when it comes to actually running a multi-step complex process with the same exact instructions on both platforms. Now, in addition, one more thing that you can do in custom GPTs that you cannot do in Gems is add code to them to connect them to more advanced capabilities and connect them to API of third-party tools, connect them to MCP servers, et cetera.

14:43So if we look at T, this particular GPT is connected down here to the API of data for SEO, which is a company that provides data for SEO, like the name suggests. But that means that I can ask questions about keywords and their availability and competitors and their density and so on across any domain that I want within this custom GPT and get information about myself, my competitors, and so on. and build my future blog post based on actual real SEO data right here within ChatGPT without having to go to other tools and bring information and so on. This is a very powerful capability. This is called Actions, and it's available on the bottom of Custom GPTs.

15:22If I open this, you'll see this is just code that connects it to their MCP server. I didn't write the code. That's a code that they provide to anybody. So you can go and grab the code and attach it in here, and you can then have this kind of capability. The ability to outcode into a gem just doesn't exist. So that's another benefit of custom GPTs versus Gemini gems. The next topic that we're going to talk about is dashboards. So I actually think that the best tool to create dashboards right now is Claude. Every time I need to create dashboards, I actually go to Claude, and Claude does an amazing job in creating visually pleasing and also very effective dashboards.

15:56However, I wanted to test Gemini on that topic. But before I even wanted to test it, something very interesting happened. I was actually working on a financial analysis. And so the financial analysis, just as an example, took information of a publicly traded company. In this particular case, I believe it was Salesforce. And then it grabbed the information from the internet and it gave me their information. And I asked it to create a report about it. So this was a great example on how you can take raw data off the internet and create graphs like you can see here, like fiscal results per quarter, or then compare that year over year and just get information just by asking simple questions or even creating complete reports like you can see here inside of Gemini.

16:33So it created this detailed report on the financial success of Salesforce in the past X number of quarters based on the prompting that I provided in the beginning. I could have done this obviously on any company in the world. So a very solid solution from Gemini in this particular case. However, it then suggested on its own to create an interactive dashboard. And all I have to do is say, yes, I'm interested in that. And it created this amazing dashboard from the data that it has collected based on my prompt. So you can see it says Salesforce Financial Dashboard. It has key matrix, revenue analysis, profitability, and efficiency.

17:03Each and every one of them is a link that if I'll click, it will scroll down to that particular section. Then there's the key findings in big, bold statuses. And if I click on each and every one of them, it opens another segment with some more details about this particular topic. So you can see if I click different things, it goes to that. And then it has revenue and seasonal patterns. And then it has year-over-year comparison. And it created all of this on its own based on the conversation that I had with it. I did not define any of the components in this dashboard. It just asked me if I would like to create one.

17:31And when I said yes, it created this, which is really, really cool. So from that perspective, huge win on the Gemini side, both taking the initiative and suggesting it, but also without me defining what exactly I want to see in the dashboard, it was able to create this amazing dashboard. But then I did something else and I went the next step. I said, I have these amazing dashboards that I've created in Cloud. Can I just replicate them in ChatGPT and in Gemini and see which one does a better job? So I started with ChatGPT. And all I did is I provided it the same exact information that I provided to Claude.

18:02And I provided it a screenshot of the dashboard. So I've got the input file that I'm going to upload, basically the same exact file that I'm using on the Claude side. And you've got a screenshot of the entire page. So a full page screenshot, not just the top of it, of the dashboard and all the components that are in it. and then I ask it and I said, you're an expert in dashboard application creation. I'm going to provide you the following, basically what I explained to you and I asked you to create the dashboard and it did and it never works. So when I click preview, the first step is correct.

18:30It actually asked me to upload the file and start the analysis. But every time we do this, there's some kind of a bug and the actual dashboard doesn't run despite the fact it's the same exact file I gave it as the sample data in order to build the initial dashboard. So that was my experience with ChatGPT. I tried it several times and I got the same issue every single time. Jem and I, on the other hand, got the same exact instructions. I literally copied and pasted the instructions. I also copied and pasted the same screenshot from the Claude dashboard. And instead, of course, I will create the dashboard.

18:59And he went ahead and created an amazing dashboard. It's working. It's fully functional. I can upload the file. So let's play the game. I will click here. And I will go and find the file. Here we go. Click open. Click start analysis. And boom, I have a working dashboard. And you can see that it has the monthly sales trend. And it has year-over-year growth. and it has the top 20 clients and their trend in the past month compared with the trailing 12 months and it has the distribution between the top leading clients and there's the filters on top and i didn't have to explain any of this i literally just showed it a screenshot of the other dashboard so if i click now channel one you can see that it changes the numbers i can click and say oh i just want to look at q1 uh for that so it will change that as well and you can see that it changes only to q1 across the different years uh and so on and so forth so all of this is working i can go to clients and i can select the different clients across different regions and everything will move and update accordingly across all the different aspects of the dashboard.

19:52Really fantastic and without almost any effort. So on this particular example, definitely a huge win for Gemini for creating the right code, for understanding the need and for creating a very useful and functioning dashboard with very few steps. The only thing I had to actually correct was the graph on the bottom that had all the companies if there are no filters are applied. So if I do this, it had like a gazillion companies, which made no sense. It was too crowded. And all I had to do is ask it to only include the top 20 companies in the graph, and everything else just worked out of the box with almost no instructions.

20:23So big win to Jim and I on this particular creation of the dashboard compared to ChatGPT that, despite all my attempts, was never able to make this work. And now our next topic is deep research. Both tools have deep research creating capabilities. Both tools are the two leading tools from a deep research perspective right now, from my perspective in my experience, and I use both of them all the time. I also try the other tools very regularly. So about once every other week, I will try one of the other tools or some of the other tools as well, just to see how they're doing. So whether that's perplexity or DeepSeek or Claude, and I will try their DeepSeek as well.

20:57Still, Chachapiti and Gemini have a very big lead when it comes to DeepSeek. But let's compare the two. And so in this particular example, what I was looking for is I was looking for AI software solutions that can help architects, interior designers, smart home integrators, and builders, and other people who work with floor plans and stuff like that, because I have multiple clients in that universe, and I wanted to see what it will find. So I wrote a long and detailed prompt. I copied the same exact prompt to both tools, and I let it run. So you can see the prompt is about a page and a half long, explaining exactly what I need.

21:26The way it works with Gemini, it tells you what it's going to do, and then you can say that you approve the plan or not. If you don't, it will ask you to clarify what else do you want or what you want different. It's a little different on the ChatUPT universe. It always asks you questions. So after you write the prompt, again, it's the same exact prompt I copied and pasted it. It said, to get started, could you please clarify the following? Are you focused only on AI-powered software or should I include any advanced software, even if not AI-based, that supports automation, rules, logic, blah, blah, blah?

21:53So it asked me all these questions that I got to answer in order to clarify exactly what I'm looking for. And then both started the research. Usually Gemini takes a lot longer when it does the process, but both tools will work on it for a while. And then this is the result I got from ChatGPT. So you can see it starts with a table that shows the different tools and what functionality they have. But what you will see is that already here there's issues where it's not showing some of the table. And if you scroll down, then you'll see many of these aspects, instead of being a part of the table as they should, are just written in text with the vertical line in between them.

22:29So I didn't really create it correctly. So I have some of the functionality here compared, but then all the other ones are in this weird format that I cannot do anything with. So I messed that up. But overall, then it started with generative design tools for architects and gave me several different options. Then it gave me AI tools for interior designer and space planning. And then it gave me a few options here and so on and so forth. So overall, a detailed, helpful report that is aligned with what I requested or kind of aligned with what I requested. And I explain why, what I mean by kind of aligned.

23:00If you go back to the prompt, what I ask it to start with, I ask it to start with doing a research on what kind of positions, what kind of roles exist in these companies and what are the use cases of each and every one of those people. And only then go and start the research on the actual tools to be able to check if they can do these functions that these different roles have to do. So if we go to the Gemini version of the deep research, you will see that it followed the instructions perfectly. It said part one, the professional landscape of floor plan utilization. And then it has objectives, and then it has chapter one, core design and planning professions.