Don’t Use AI for Slides Until You Watch This

In today's video, we'll go through the concepts which you need to know if you are using any of the AI tools to create PowerPoint presentations, to create nice slide decks, or in general, if you're using the tools which are saying that they use AI. In order to understand how AI tools work, first of all, we need to understand a couple of concepts. And we will start with the following one. That the model, the large language model, which is processing and creating your slides, is not living on your laptop.

What do I mean? So let's imagine that you are having a work laptop, and you are sending a request to such tools as Beautiful AI, Gamma AI, Claude, ChatGPT, or any other tool which is creating slides. I'm just using the the most often used ones here. So in order for you to get back the slides, these companies should send your data. Normally, it's your prompt or maybe some other documents which are attached to the servers where large language models are running. So it looks like that.

You are sending from your workload... Laptop some kind of prompt. Let's put this prompt here. These companies are taking this information, and then they are sending it further. So let's say they're sending somewhere here to this large language model. And by the way, the large language model can be on the premises of these companies like in Claude or ChatGPT or Beautiful AI and Gamma AI. They will use so called subprocessors. They will use the large language models provided by other companies. And let's let's, by the way, immediately look at it.

So if you if you go to the Gamma AI list of subprocessors, it's actually quite huge. You will be able to see here that there is a a whole set of tools in the category artificial intelligence from different countries, OpenAI, Perplexity, Microsoft, Luma AI, and so on. And the the list is actually quite huge. The same for Beautiful AI. It's actually a lot smaller list. It's only OpenAI and Anthropic and Google. But for AI services, these companies are using these other providers. So if we go go back here, that's what you will be able to see.

So your prompt is sent to these companies, and then these companies are sending the data to the servers where large language models are stored and where the inference is happening. So for you to get back PowerPoint... So let's imagine they basically created the PowerPoint. So first of all, they got some information from large language model, and then they send this PowerPoint back to you to you. And you receive it on your work laptop. So the concept number one, and I just saw a number of my colleagues who who were thinking that their models are actually running on their laptops.

It's not the case. They are not even running inside such companies as Beautiful AI or Gamma AI because they are using other companies which are provided... Providing large language models. So this is the first one about the model is not on your laptop. Another one... Another concept is about the company approved environment. So if you take this box and let's put it here, I will move the laptop into the company approved environment because it's a work laptop. And let's say inside the company you are using, for instance, Copilot Copilot for work, It's approved by the organization, and you're having different types of, let's say, documents, for instance.

There are some files. There is something else. There is different PowerPoint documents and other documents and prompt. Prompts with the sensitive data. They are all within this perimeter of... Which we'll company approved environment. So when you are working in an organization, it's very important for you to understand really well, when you're using AI, which information is leaving company approved environment and when, so that you will not, by mistake, send some sensitive or confidential data outside of this perimeter. And normally, what happens if these companies... And here, for instance, when I'm speaking about Claude or ChatGPT, of course, there is a ChatGPT business and a Claude business, and they may be within the company environment.

But even if we are speaking about Copilot, there are also some kind of personal personal accounts of Copilot, which lives outside of the company approved environment. So what is happening? And by the way, again, Beautiful and Gamma AI can be approved within your company to use with the sensitive data. And in this case, these two these two companies, Beautiful AI and Gamma AI, they will be living within your company approved environment. This is just a schematic example. What is important to understand here is this distinction between the company approved environment and all the data which is going inside inside of this perimeter and everything which is happening outside.

The next concept which I want to touch on is the distinction between one hop tools like Claude and ChatGPT. I already mentioned a bit about it. And two hop tools like Gamma AI and Beautiful AI. So there is a number of companies, AI companies. Actually, a lot of these companies. They are most of them do not have their own AI inside the company because it's just difficult to develop. And big companies like Anthropic, OpenAI, Google, and others, they are more suited to develop amazing large language models.

So these companies, the smaller the smaller organizations or the ones which do not want to be involved in the development of these big models, they are using the subprocessors which I have just shown you. So the Gamma AI has its own subprocessors. The Beautiful AI has subprocessors as well. So we can say that beautiful AI and gamma AI, they are two hop tools, this one. Two hop tools, which means that when you are sending the prompt, the prompt first goes to beautiful AI, and then it goes to the large language model in maybe Anthropic, maybe in another company, maybe it's...

These are the companies which work with vision, with with audio, and so on. So they are two hop two hops up... Two hop companies. So first, your information to create a slide deck is coming to this company, and then this company moves your data somewhere else. Then this somewhere else, so large language model, another provider, is given back to this company an answer, and the company is sending you back a PowerPoint slide or slide in any other format. While one hop tools like Claude and ChatGPT, they actually themselves have they have these large language models capacities and servers on their premises, and that's why the the trip of the data from your laptop to these companies, it's...

Basically, it stops there. So if you go... If you take, as an example, a Claude or GPT, your prompt will first arrive to Claude or to ChatGPT, then it's gonna be processed there internally within the infrastructure, and the resulting document, like PowerPoint presentation, comes back to your work laptop immediately from them. So I would say if you are dealing with these companies or if you're dealing with, yeah, bigger companies, which has its own infrastructure, tools, and capacity to create slides without referring to subprocessors, then in the end, legally, even it's easier for your company to approve such tools because you're just approving them.

And then once the approval is there, you do not need to think about the subprocessors, how risky they are or anything like that. Another thing is this concept about... And we'll take a look at Copilot, for example. So same logo, different place, Copilot. Work laptop doesn't matter. Account does. So what what does it mean? It means the following. So if you look at Copilot, there are different versions of Copilot. And, again, it's the same for ChatGPT and for Claude. Work version of Copilot and the personal version of Copilot.

And they are not... Even though the name is the same, but these are completely different product, and they treat the data in a completely different way. That's why even if you are working from the work laptop, but by mistake, for instance, in the browser, you opened Copilot in your personal account, then basically all the data which you'll be pasting there will be leaving the company approved environment, And normally, you will be basically violating the the rules of this organization because the sensitive data may leave to Copilot.

Even though it's Microsoft, but the data is protected in completely different way there. For example, the question of training, and it's gonna be the next one, may be completely different. In the, Copilot for Work, the data is not used to train their models. While in the personal account, the data can be used to train the models. Coming, so as a result, like, the conclusion is just the following. It's not about the name, not about the logo. It's about the type of account. It's also not about whether you're using work laptop or private laptop.

It's just into which account you are actually uploading your information and pasting the prompts. So the next one is training. So let's look at training. It's just one aspect. What does it mean? I have heard from a number of my friends and colleagues the following that they're saying that they can use or they're thinking that they can use the tool because the tool is promising them not to use their data for training. So training, the problem is just one of the elements and one of the questions.

Even if some of these companies are telling you that they will not use the data for training, let's say they're saying the data will be not used for training, it doesn't mean that this data will not be used in some other way. Because in the end, there is the question not only about whether the AI model will be will be trained on your data. Another question, which was always there before AI models appeared, what about data processing? What about storage? So if you are sending the data outside of your company approved environment to these tools or to any other tool, the question is, okay.

What's happening with the data? Where it is stored? In which servers? Is it sent to subprocessors? Is it sent to any other companies? What's the policy in terms of when this data will be deleted? Will it stay forever there? So there is a big number of questions which are there besides training. So training is just one of them. And, the last one is why it is important for slide decks. Because the video is about slide decks. It's not just about any AI tool. It is important because you should understand that creation of the slide deck with the help of AI requires AI.

And the question is where this AI is living, where it's sitting. If it is living inside the company approved environment... And by the way, here, it doesn't mean that it should be physically inside the company or on the service of the company. It can be the same company Beautiful AI, but let's say it's within the company approved environment because the company, its security team said, okay, we are fine with Beautiful AI. It's approved. You can use sensitive data with Beautiful AI. So if it happens, this way, then Beautiful AI is within the company approved environment.

Important thing to understand that if you are pasting some prompts within the company approved environment or touching some data, and then AI generates in the end something which is then converted into PowerPoint presentation, and you are very happy you're having PowerPoint presentation, then you have not done anything wrong. You basically just used approved tools. You pasted there some sensitive internal data. Everything is fine. If there is no information leakage, you have done everything in line with the requirements of your organization. But if you are using tools which are not within the perimeter of a company approved environment, it's not within the perimeter of the environment approved by your security team, then what is happening?

All your prompts which you are sending to these tools to create nice slide decks, they are actually leaving the company approved environment. And then you are receiving back the PowerPoint presentation. Amazing slides. Looks nice. But the question is, what happened with this data based on which the slides have been created? Is it there still within gamma.ai or its subprocessors? Is it with Anthropic? Is it with Charge ChatGPT or with or with OpenAI? Are we fine that our data is sitting there? And if the answer is no, then it's super important for you to understand how you should use...

If the private AI, personal AI is approved for use or it can be used or, let's say, not prohibited to use within your company, how can you responsibly use these tools? When it happens that the data is... Sensitive data is leaking to these tools, and how to avoid this leakage. If you like this explanation, I would appreciate if you subscribe to this channel, leave the comment below, and also check the longer video, which will appear in the next days, where you will get an explanation how to use your personal AI tools in a responsible way so that your work environment and your work experience just benefits from the fact that you have subscription to these tools in a private life.

Thank you for watching, and I'll see you in the next video. Bye.