WEBVTT
00:00:22.929 --> 00:00:33.009
Hello, and welcome to the Community IT Innovators Technology Topics podcast, nonprofit AI Midweek Check-in.
00:00:33.009 --> 00:00:53.009
I'm Carolyn Woodard, your host, and today I wanted to share kind of a shocking, maybe not shocking, story about the head, the interim director of the Cybersecurity and Infrastructure Security Agency in the United States government, which is known by its acronym of CISA.
00:00:54.209 --> 00:01:11.890
And in this podcast and on our webinars, we are very careful to tell our listeners to be very careful at using public, quote unquote public tools like ChatGPT, where you go out to just a website and you are just a consumer.
00:01:11.890 --> 00:01:22.849
You have no, you might have to make an account with them, but if you look at the terms and service, uh they you're they are using your materials to train their tools.
00:01:22.849 --> 00:01:26.530
And that means they have access to everything that you upload.
00:01:26.530 --> 00:01:29.329
And, you know, for various reasons.
00:01:29.329 --> 00:01:42.769
I mean, there are some things that, you know, I would trust Chat GPT with, like if you're looking for a recipe or you're trying to find some information on restaurants or something that you don't just don't really care about it being private.
00:01:42.769 --> 00:01:56.609
Um, but certainly for anything related to your uh job at your nonprofit, we are highly, highly recommending that people use the enterprise version of the tool that they want to use.
00:01:56.689 --> 00:02:00.209
And that just means have you know pay for the enterprise license.
00:02:00.209 --> 00:02:12.210
If you're on Microsoft Office 365 or Google Workspace, you probably already have a license associated with your enterprise subscriptions.
00:02:12.210 --> 00:02:17.330
So if you don't know how to access it, you can talk to people at your nonprofit.
00:02:17.490 --> 00:02:22.129
We're going to be having the webinar in February, on February 25th.
00:02:22.129 --> 00:02:36.049
Our CTO and cybersecurity expert, Matt Eschelman, is going to go through in some detail how do you know if you're signing on to your company account and just make sure that you're doing that.
00:02:36.049 --> 00:02:38.849
Um, so but it's something that we talk a lot about.
00:02:38.849 --> 00:02:40.769
Like you may have some confusion about it.
00:02:40.930 --> 00:02:59.409
It's it can be confusing, but one would not expect the uh acting cyber chief to upload sensitive files to a public version of ChatGPT, but it came out uh last week in Politico and some other uh news outlets have covered it.
00:02:59.409 --> 00:03:01.969
That that's in fact exactly what happened.
00:03:01.969 --> 00:03:03.889
It was over last summer.
00:03:03.889 --> 00:03:19.409
Um the interim head of the cyber defense agency uploaded sensitive contracting documents into a public version of Chat GPT, triggering multiple automated security warnings that are meant to stop the theft or unintentional disclosure of government material.
00:03:19.409 --> 00:03:34.609
And uh, this was Madu Gottum ukkala, um, who is the acting director of cybersecurity and infrastructure security agency, and he had actually requested special permission to use Chat GPT.
00:03:34.609 --> 00:03:46.449
And one is just surprised that there was any kind of confusion about what he was allowed to upload to the public um ChatGPT or not.
00:03:46.449 --> 00:03:50.930
But in any case, I will share that uh link with you in the show notes.
00:03:50.930 --> 00:03:55.889
Um, this should hopefully just be a reminder that you know it can happen.
00:03:55.889 --> 00:04:01.329
It shouldn't be happening at that level, but it's definitely something to be concerned about.
00:04:01.810 --> 00:04:23.569
Um, so just to go over again, if you um Google or search for the difference between Microsoft Copilot and OpenAI's Chat GPT, they're they're essentially the same tool, but the differences uh center on your data privacy and the business integration.
00:04:23.569 --> 00:04:25.730
So you get what you pay for.
00:04:25.730 --> 00:04:33.250
Uh generally, Copilot with a 365 subscription is designed for enterprise level data security.
00:04:33.250 --> 00:04:37.569
Uh whereas, you know, Chat GPT, it's free.
00:04:37.569 --> 00:04:46.050
Like you are uh how they're making their or going to make their money eventually is by using your data and your information.
00:04:46.050 --> 00:04:48.930
Um and that's why it's quote unquote free.
00:04:49.250 --> 00:05:00.370
So if you're using Copilot, you have data protection, you can you know look into all of this for yourself and uh look into the terms and conditions agreement that you have.
00:05:00.370 --> 00:05:04.050
You can ask at your nonprofit about it.
00:05:04.050 --> 00:05:12.610
For data protection, your prompts and data are not used to train the underlying large language model for copilot.
00:05:12.610 --> 00:05:17.250
It keeps your data within your enterprise cloud boundary.
00:05:17.250 --> 00:05:20.930
It has commercial data protection limitations.
00:05:20.930 --> 00:05:29.490
So this protection of your data only applies when you're signed in to an eligible work or school account.
00:05:29.490 --> 00:05:35.170
It may have some access limitations, which is good and bad.
00:05:35.329 --> 00:05:36.529
We've talked about this a little bit.
00:05:36.529 --> 00:05:46.930
We've been talking a lot about trying to organize and clean up your data now that there is AI search available.
00:05:46.930 --> 00:05:59.250
So if you have documents that have been shared or they're in a folder that is shared, in the past, you might have what's called privacy by obscurity.
00:05:59.250 --> 00:06:05.490
So, unless you went looking for those documents, you wouldn't come across them.
00:06:05.490 --> 00:06:11.329
And who really has the time to go around and poke around in different folders that you might have access to?
00:06:11.329 --> 00:06:13.649
It was just something that was kind of difficult.
00:06:13.649 --> 00:06:21.170
So there was this kind of layer of protection that just it was hard to access some of the things that were shared in some of those folders.
00:06:21.170 --> 00:06:28.449
With AI search, you can find any document that you have permissions to.
00:06:28.449 --> 00:06:31.329
And that may not be you specifically.
00:06:31.329 --> 00:06:45.810
It may be that the person who owns that document sent the permissions to anyone with this link can find this document, in which case your Copilot is going to see that and provide you with access to it.
00:06:45.810 --> 00:06:56.529
If the terms that you're looking for are in that document, it's gonna surface that for you and um show it to you because you have act because you have you have access to it.
00:06:56.529 --> 00:07:04.050
There are ways that you can go through your files and look for files that are set that way with these open uh access.
00:07:04.689 --> 00:07:13.810
Um, but the point for this conversation is that Copilot can only access and use organizational data that the user has permission to see.
00:07:13.810 --> 00:07:29.569
So it's like good in that if you have like your HR folders set up correctly in SharePoint, uh, then the random user at your organization should not be able to find those.
00:07:29.569 --> 00:07:37.250
So Copilot can only access and use organizational data that the user already has permission to see.
00:07:37.250 --> 00:07:45.329
So that should protect sensitive documents that are not shared or that are in folders that are not shared.
00:07:45.409 --> 00:07:59.089
But again, that's something that you're going to have to undertake as a way to clean up your data, protect your data, and ensure that you have clean and protected data going forward.
00:07:59.089 --> 00:08:11.649
It's something that you're going to need to do staff training on and just continue, like, don't assume that if you've done the staff training once on how you uh save or share a document, then it'll stick.
00:08:11.649 --> 00:08:23.490
You need to be uh revisiting, you know, quarterly, probably or more frequently, um, that the AI search is turning up everything that has quote unquote open permissions.
00:08:23.490 --> 00:08:27.250
So make sure that you're um staying on top of that.
00:08:27.490 --> 00:08:35.730
Other Microsoft Copilot terms uh for the enterprise version is that deep fakes and harassment are prohibited.
00:08:35.730 --> 00:08:43.330
The terms of use prohibit creating content to impersonate people, engage in doxing, or create adult content or violence.
00:08:43.330 --> 00:08:45.009
Of course, it's a work account, right?
00:08:45.009 --> 00:08:49.889
So hopefully one is not doing those types of activities while at work.
00:08:49.889 --> 00:09:07.409
And I would say this is one of those terms and conditions where like you could do it, but then when you were caught, you would be in violation of both the Microsoft Copilot terms and probably your own user acceptable use policies at your organization.
00:09:07.409 --> 00:09:11.169
So that is a level of protection that's there.
00:09:11.330 --> 00:09:13.250
And then copyright commitment.
00:09:13.250 --> 00:09:26.610
Um, Microsoft has expanded its customer copyright commitment to cover commercial customers, promising to defend them against copyright infringement claims for using copilot outputs, but providing proper guardrails were used.
00:09:26.610 --> 00:09:33.090
So again, you're gonna want to check those terms and also check your acceptable use policies at your organization.
00:09:33.330 --> 00:09:37.889
For open AI Chat GPT, it's pretty much the opposite on everything.
00:09:37.889 --> 00:09:45.730
So by default, uh Chat GPT may use your conversations, your inputs, anything you upload to it to train their models.
00:09:45.730 --> 00:09:51.009
You can opt out of that data use, but it will limit your functionality.
00:09:51.009 --> 00:09:59.009
The ownership of the output, uh, you own the output, but you must not represent it as human-generated if it was not.
00:09:59.009 --> 00:10:07.970
So I think it's kind of squishy that you have to acknowledge it was AI, but you can't like create an artwork and say, I did this without AI.
00:10:07.970 --> 00:10:10.129
That would go against their terms of service.
00:10:10.370 --> 00:10:15.330
The terms state that you must not rely on ChatGPT for professional, legal, or medical advice.
00:10:15.330 --> 00:10:29.009
So again, one of the squishy things, like you can find that advice out there, but if you use it, don't come after ChatGPT because they've already said don't use this as a as acceptable medical or legal advice.
00:10:29.009 --> 00:10:35.570
Similar to Copilot, uh, ChatGPT prohibits using the service to generate disinformation, deceive, or harm others.
00:10:35.570 --> 00:10:46.850
And I'm gonna say right now a big ha that they will come after you for that because as we know, it is being used for those purposes as we speak.
00:10:46.850 --> 00:10:52.049
All right, so that is the difference for copilot versus chat GPT.
00:10:52.370 --> 00:11:01.169
For Google, it's a little bit more confusing because they call it Gemini in both instances, but it's very, very similar differences.
00:11:01.169 --> 00:11:05.490
So the main difference is between the terms of service for the public Gemini.
00:11:05.490 --> 00:11:15.570
So you just go out and you, you know, Google Gemini and you sign into it and you use it, you ask it questions or upload things to it for help.
00:11:15.570 --> 00:11:26.689
And the Gemini for Google workspace for the enterprise and business, um, the big differences again are data privacy, the model training, and the intellectual property protection.
00:11:26.689 --> 00:11:34.929
And they say public Gemini is for personal use, enterprise Gemini is for business use with enterprise grade protections.
00:11:35.090 --> 00:11:40.529
Again, the big differences in the terms of service include data privacy and model training.
00:11:40.529 --> 00:11:48.450
If you're using the public Gemini, Google may use your conversations, prompts, and feedbacks to train their AI models and improve services.
00:11:48.450 --> 00:11:52.129
You can turn that off as you can with Chat GPT.
00:11:52.129 --> 00:11:58.689
Some data is kept for 72 hours, and um, yeah, it will limit your functionality.
00:11:58.689 --> 00:12:08.689
They're relying on you to be uploading prompts and questions and documents so that it can learn even better how to respond to you.
00:12:08.689 --> 00:12:14.929
For the Enterprise Gemini, when you're using workspace, your data is not used to train or improve Gemini models.
00:12:14.929 --> 00:12:16.689
It is not reviewed by humans.
00:12:16.689 --> 00:12:24.850
Prompts and generated content are not shared with other customers, uh, with the exception, which some nonprofits may be interested in this.
00:12:24.850 --> 00:12:31.730
I did not realize this, but there is a starter edition of Workspace for a 30-day trial.
00:12:31.730 --> 00:12:37.490
And if you're using Gemini during that 30 days, uh it says that it may use your data for training.
00:12:37.490 --> 00:12:48.689
So again, once you're in the fully paid uh subscription enterprise license, your data is kept within your organization.
00:12:48.689 --> 00:13:02.370
I guess the way to explain it is that it is it is still using its model to respond to you, but your things that you're uploading to it are not going out and becoming part of that model.
00:13:03.169 --> 00:13:13.730
So a difference in intellectual property and content ownership when you're using public Gemini, the ownership of that generated content is very complex.
00:13:13.730 --> 00:13:16.370
I'm not even going to begin to go into it.
00:13:16.370 --> 00:13:22.929
In terms of Enterprise Gemini, so like for copyright and such, you own your data, not Google.
00:13:22.929 --> 00:13:32.769
Generated output from Enterprise Gemini is considered to be customer data, meaning your AI-generated assets belong to your organization.
00:13:33.169 --> 00:13:38.210
Data residency and compliance, um, public Gemini, they just have it.
00:13:38.210 --> 00:13:40.370
Like you don't store it anymore.
00:13:40.370 --> 00:13:52.769
For Enterprise Gemini, it supports compliance with HIPAA, GDPR for Europe, and FedRAMP, which is what you need to be compliant with to work with federal government in the United States.
00:13:52.769 --> 00:13:58.370
It supports data residency policies and lets administrators control where data is stored.
00:13:58.370 --> 00:14:00.370
And again, like those permissions.
00:14:00.610 --> 00:14:02.049
So data scope.
00:14:02.049 --> 00:14:08.689
With public Gemini, it is limited to public web data and input from the current session.
00:14:08.689 --> 00:14:11.809
So you can only use things that are publicly available.
00:14:11.809 --> 00:14:17.570
With Enterprise Gemini, it operates within your secure private ecosystem.
00:14:17.570 --> 00:14:24.129
It can use your organization's data, internal documents, Gmail, anything on Drive.
00:14:24.129 --> 00:14:28.929
It can also connect to third-party data sources, so like your Salesforce.
00:14:28.929 --> 00:14:31.169
So that's something to consider.
00:14:31.649 --> 00:14:37.250
And then security controls, if you're using public Gemini, of course, there's no administrator.
00:14:37.250 --> 00:14:39.169
You're just out there in the public.
00:14:39.169 --> 00:14:41.169
So you have no control.
00:14:41.169 --> 00:14:54.610
For enterprise Gemini, it does have administrative controls, so IT can manage access, and that allows it to support data loss prevention and certain types of encryptions.
00:14:54.610 --> 00:15:08.529
So those are the quick uh updates on if you were confused about the differences between public and private AI tools for the two big players, Microsoft and Google.
00:15:08.529 --> 00:15:10.850
I just wanted to quickly go over it.
00:15:11.169 --> 00:15:28.529
If you're interested in other tools like Claude, Anthropics Claude, or Perplexity or Change AI, and you would want to just be aware, if you're creating a free account, you're essentially using the public version.
00:15:28.529 --> 00:15:39.250
If you purchase a subscription either for yourself through your workplace or for your workplace and enterprise version, then you have those additional protections.
00:15:39.250 --> 00:15:42.929
Your data is private to your organization.
00:15:42.929 --> 00:15:45.490
It's still super, super functional.
00:15:45.490 --> 00:15:47.649
So I want to emphasize that.
00:15:47.649 --> 00:16:06.210
It's just that, especially if you are concerned, you have sensitive information, you just want to protect your privacy and the privacy of your data, which of course everyone does, we highly strongly recommend to use the enterprise version of the AI tool that you're interested in.
00:16:07.169 --> 00:16:34.289
Of course, if you are using a tool that is updating you with AI, so for example, Zoom, uh Zapier, maybe your CRM, Salesforce, if they have AI and you like do the update and you now you have AI, that is also going to, for the most part, be controlled within your enterprise license for that tool for Zoom or Zapier or Salesforce.
00:16:34.289 --> 00:16:43.250
But you know, it's probably good practice to continue to look at the terms and conditions and just make sure that you're using AI safely at work.
00:16:43.250 --> 00:16:44.850
All right.
00:16:45.009 --> 00:16:49.649
Well, that's it for this episode, midweek check in on nonprofit AI.
00:16:49.649 --> 00:16:54.289
And I'll see you again on Friday with our regularly scheduled podcast.
00:16:54.289 --> 00:16:59.889
And again, every Tuesday we'll be coming back with some quick AI tips for nonprofits.
00:16:59.889 --> 00:17:00.769
Take care.
00:00:22.929 --> 00:00:33.009
Hello, and welcome to the Community IT Innovators Technology Topics podcast, nonprofit AI Midweek Check-in.
00:00:33.009 --> 00:00:53.009
I'm Carolyn Woodard, your host, and today I wanted to share kind of a shocking, maybe not shocking, story about the head, the interim director of the Cybersecurity and Infrastructure Security Agency in the United States government, which is known by its acronym of CISA.
00:00:54.209 --> 00:01:11.890
And in this podcast and on our webinars, we are very careful to tell our listeners to be very careful at using public, quote unquote public tools like ChatGPT, where you go out to just a website and you are just a consumer.
00:01:11.890 --> 00:01:22.849
You have no, you might have to make an account with them, but if you look at the terms and service, uh they you're they are using your materials to train their tools.
00:01:22.849 --> 00:01:26.530
And that means they have access to everything that you upload.
00:01:26.530 --> 00:01:29.329
And, you know, for various reasons.
00:01:29.329 --> 00:01:42.769
I mean, there are some things that, you know, I would trust Chat GPT with, like if you're looking for a recipe or you're trying to find some information on restaurants or something that you don't just don't really care about it being private.
00:01:42.769 --> 00:01:56.609
Um, but certainly for anything related to your uh job at your nonprofit, we are highly, highly recommending that people use the enterprise version of the tool that they want to use.
00:01:56.689 --> 00:02:00.209
And that just means have you know pay for the enterprise license.
00:02:00.209 --> 00:02:12.210
If you're on Microsoft Office 365 or Google Workspace, you probably already have a license associated with your enterprise subscriptions.
00:02:12.210 --> 00:02:17.330
So if you don't know how to access it, you can talk to people at your nonprofit.
00:02:17.490 --> 00:02:22.129
We're going to be having the webinar in February, on February 25th.
00:02:22.129 --> 00:02:36.049
Our CTO and cybersecurity expert, Matt Eschelman, is going to go through in some detail how do you know if you're signing on to your company account and just make sure that you're doing that.
00:02:36.049 --> 00:02:38.849
Um, so but it's something that we talk a lot about.
00:02:38.849 --> 00:02:40.769
Like you may have some confusion about it.
00:02:40.930 --> 00:02:59.409
It's it can be confusing, but one would not expect the uh acting cyber chief to upload sensitive files to a public version of ChatGPT, but it came out uh last week in Politico and some other uh news outlets have covered it.
00:02:59.409 --> 00:03:01.969
That that's in fact exactly what happened.
00:03:01.969 --> 00:03:03.889
It was over last summer.
00:03:03.889 --> 00:03:19.409
Um the interim head of the cyber defense agency uploaded sensitive contracting documents into a public version of Chat GPT, triggering multiple automated security warnings that are meant to stop the theft or unintentional disclosure of government material.
00:03:19.409 --> 00:03:34.609
And uh, this was Madu Gottum ukkala, um, who is the acting director of cybersecurity and infrastructure security agency, and he had actually requested special permission to use Chat GPT.
00:03:34.609 --> 00:03:46.449
And one is just surprised that there was any kind of confusion about what he was allowed to upload to the public um ChatGPT or not.
00:03:46.449 --> 00:03:50.930
But in any case, I will share that uh link with you in the show notes.
00:03:50.930 --> 00:03:55.889
Um, this should hopefully just be a reminder that you know it can happen.
00:03:55.889 --> 00:04:01.329
It shouldn't be happening at that level, but it's definitely something to be concerned about.
00:04:01.810 --> 00:04:23.569
Um, so just to go over again, if you um Google or search for the difference between Microsoft Copilot and OpenAI's Chat GPT, they're they're essentially the same tool, but the differences uh center on your data privacy and the business integration.
00:04:23.569 --> 00:04:25.730
So you get what you pay for.
00:04:25.730 --> 00:04:33.250
Uh generally, Copilot with a 365 subscription is designed for enterprise level data security.
00:04:33.250 --> 00:04:37.569
Uh whereas, you know, Chat GPT, it's free.
00:04:37.569 --> 00:04:46.050
Like you are uh how they're making their or going to make their money eventually is by using your data and your information.
00:04:46.050 --> 00:04:48.930
Um and that's why it's quote unquote free.
00:04:49.250 --> 00:05:00.370
So if you're using Copilot, you have data protection, you can you know look into all of this for yourself and uh look into the terms and conditions agreement that you have.
00:05:00.370 --> 00:05:04.050
You can ask at your nonprofit about it.
00:05:04.050 --> 00:05:12.610
For data protection, your prompts and data are not used to train the underlying large language model for copilot.
00:05:12.610 --> 00:05:17.250
It keeps your data within your enterprise cloud boundary.
00:05:17.250 --> 00:05:20.930
It has commercial data protection limitations.
00:05:20.930 --> 00:05:29.490
So this protection of your data only applies when you're signed in to an eligible work or school account.
00:05:29.490 --> 00:05:35.170
It may have some access limitations, which is good and bad.
00:05:35.329 --> 00:05:36.529
We've talked about this a little bit.
00:05:36.529 --> 00:05:46.930
We've been talking a lot about trying to organize and clean up your data now that there is AI search available.
00:05:46.930 --> 00:05:59.250
So if you have documents that have been shared or they're in a folder that is shared, in the past, you might have what's called privacy by obscurity.
00:05:59.250 --> 00:06:05.490
So, unless you went looking for those documents, you wouldn't come across them.
00:06:05.490 --> 00:06:11.329
And who really has the time to go around and poke around in different folders that you might have access to?
00:06:11.329 --> 00:06:13.649
It was just something that was kind of difficult.
00:06:13.649 --> 00:06:21.170
So there was this kind of layer of protection that just it was hard to access some of the things that were shared in some of those folders.
00:06:21.170 --> 00:06:28.449
With AI search, you can find any document that you have permissions to.
00:06:28.449 --> 00:06:31.329
And that may not be you specifically.
00:06:31.329 --> 00:06:45.810
It may be that the person who owns that document sent the permissions to anyone with this link can find this document, in which case your Copilot is going to see that and provide you with access to it.
00:06:45.810 --> 00:06:56.529
If the terms that you're looking for are in that document, it's gonna surface that for you and um show it to you because you have act because you have you have access to it.
00:06:56.529 --> 00:07:04.050
There are ways that you can go through your files and look for files that are set that way with these open uh access.
00:07:04.689 --> 00:07:13.810
Um, but the point for this conversation is that Copilot can only access and use organizational data that the user has permission to see.
00:07:13.810 --> 00:07:29.569
So it's like good in that if you have like your HR folders set up correctly in SharePoint, uh, then the random user at your organization should not be able to find those.
00:07:29.569 --> 00:07:37.250
So Copilot can only access and use organizational data that the user already has permission to see.
00:07:37.250 --> 00:07:45.329
So that should protect sensitive documents that are not shared or that are in folders that are not shared.
00:07:45.409 --> 00:07:59.089
But again, that's something that you're going to have to undertake as a way to clean up your data, protect your data, and ensure that you have clean and protected data going forward.
00:07:59.089 --> 00:08:11.649
It's something that you're going to need to do staff training on and just continue, like, don't assume that if you've done the staff training once on how you uh save or share a document, then it'll stick.
00:08:11.649 --> 00:08:23.490
You need to be uh revisiting, you know, quarterly, probably or more frequently, um, that the AI search is turning up everything that has quote unquote open permissions.
00:08:23.490 --> 00:08:27.250
So make sure that you're um staying on top of that.
00:08:27.490 --> 00:08:35.730
Other Microsoft Copilot terms uh for the enterprise version is that deep fakes and harassment are prohibited.
00:08:35.730 --> 00:08:43.330
The terms of use prohibit creating content to impersonate people, engage in doxing, or create adult content or violence.
00:08:43.330 --> 00:08:45.009
Of course, it's a work account, right?
00:08:45.009 --> 00:08:49.889
So hopefully one is not doing those types of activities while at work.
00:08:49.889 --> 00:09:07.409
And I would say this is one of those terms and conditions where like you could do it, but then when you were caught, you would be in violation of both the Microsoft Copilot terms and probably your own user acceptable use policies at your organization.
00:09:07.409 --> 00:09:11.169
So that is a level of protection that's there.
00:09:11.330 --> 00:09:13.250
And then copyright commitment.
00:09:13.250 --> 00:09:26.610
Um, Microsoft has expanded its customer copyright commitment to cover commercial customers, promising to defend them against copyright infringement claims for using copilot outputs, but providing proper guardrails were used.
00:09:26.610 --> 00:09:33.090
So again, you're gonna want to check those terms and also check your acceptable use policies at your organization.
00:09:33.330 --> 00:09:37.889
For open AI Chat GPT, it's pretty much the opposite on everything.
00:09:37.889 --> 00:09:45.730
So by default, uh Chat GPT may use your conversations, your inputs, anything you upload to it to train their models.
00:09:45.730 --> 00:09:51.009
You can opt out of that data use, but it will limit your functionality.
00:09:51.009 --> 00:09:59.009
The ownership of the output, uh, you own the output, but you must not represent it as human-generated if it was not.
00:09:59.009 --> 00:10:07.970
So I think it's kind of squishy that you have to acknowledge it was AI, but you can't like create an artwork and say, I did this without AI.
00:10:07.970 --> 00:10:10.129
That would go against their terms of service.
00:10:10.370 --> 00:10:15.330
The terms state that you must not rely on ChatGPT for professional, legal, or medical advice.
00:10:15.330 --> 00:10:29.009
So again, one of the squishy things, like you can find that advice out there, but if you use it, don't come after ChatGPT because they've already said don't use this as a as acceptable medical or legal advice.
00:10:29.009 --> 00:10:35.570
Similar to Copilot, uh, ChatGPT prohibits using the service to generate disinformation, deceive, or harm others.
00:10:35.570 --> 00:10:46.850
And I'm gonna say right now a big ha that they will come after you for that because as we know, it is being used for those purposes as we speak.
00:10:46.850 --> 00:10:52.049
All right, so that is the difference for copilot versus chat GPT.
00:10:52.370 --> 00:11:01.169
For Google, it's a little bit more confusing because they call it Gemini in both instances, but it's very, very similar differences.
00:11:01.169 --> 00:11:05.490
So the main difference is between the terms of service for the public Gemini.
00:11:05.490 --> 00:11:15.570
So you just go out and you, you know, Google Gemini and you sign into it and you use it, you ask it questions or upload things to it for help.
00:11:15.570 --> 00:11:26.689
And the Gemini for Google workspace for the enterprise and business, um, the big differences again are data privacy, the model training, and the intellectual property protection.
00:11:26.689 --> 00:11:34.929
And they say public Gemini is for personal use, enterprise Gemini is for business use with enterprise grade protections.
00:11:35.090 --> 00:11:40.529
Again, the big differences in the terms of service include data privacy and model training.
00:11:40.529 --> 00:11:48.450
If you're using the public Gemini, Google may use your conversations, prompts, and feedbacks to train their AI models and improve services.
00:11:48.450 --> 00:11:52.129
You can turn that off as you can with Chat GPT.
00:11:52.129 --> 00:11:58.689
Some data is kept for 72 hours, and um, yeah, it will limit your functionality.
00:11:58.689 --> 00:12:08.689
They're relying on you to be uploading prompts and questions and documents so that it can learn even better how to respond to you.
00:12:08.689 --> 00:12:14.929
For the Enterprise Gemini, when you're using workspace, your data is not used to train or improve Gemini models.
00:12:14.929 --> 00:12:16.689
It is not reviewed by humans.
00:12:16.689 --> 00:12:24.850
Prompts and generated content are not shared with other customers, uh, with the exception, which some nonprofits may be interested in this.
00:12:24.850 --> 00:12:31.730
I did not realize this, but there is a starter edition of Workspace for a 30-day trial.
00:12:31.730 --> 00:12:37.490
And if you're using Gemini during that 30 days, uh it says that it may use your data for training.
00:12:37.490 --> 00:12:48.689
So again, once you're in the fully paid uh subscription enterprise license, your data is kept within your organization.
00:12:48.689 --> 00:13:02.370
I guess the way to explain it is that it is it is still using its model to respond to you, but your things that you're uploading to it are not going out and becoming part of that model.
00:13:03.169 --> 00:13:13.730
So a difference in intellectual property and content ownership when you're using public Gemini, the ownership of that generated content is very complex.
00:13:13.730 --> 00:13:16.370
I'm not even going to begin to go into it.
00:13:16.370 --> 00:13:22.929
In terms of Enterprise Gemini, so like for copyright and such, you own your data, not Google.
00:13:22.929 --> 00:13:32.769
Generated output from Enterprise Gemini is considered to be customer data, meaning your AI-generated assets belong to your organization.
00:13:33.169 --> 00:13:38.210
Data residency and compliance, um, public Gemini, they just have it.
00:13:38.210 --> 00:13:40.370
Like you don't store it anymore.
00:13:40.370 --> 00:13:52.769
For Enterprise Gemini, it supports compliance with HIPAA, GDPR for Europe, and FedRAMP, which is what you need to be compliant with to work with federal government in the United States.
00:13:52.769 --> 00:13:58.370
It supports data residency policies and lets administrators control where data is stored.
00:13:58.370 --> 00:14:00.370
And again, like those permissions.
00:14:00.610 --> 00:14:02.049
So data scope.
00:14:02.049 --> 00:14:08.689
With public Gemini, it is limited to public web data and input from the current session.
00:14:08.689 --> 00:14:11.809
So you can only use things that are publicly available.
00:14:11.809 --> 00:14:17.570
With Enterprise Gemini, it operates within your secure private ecosystem.
00:14:17.570 --> 00:14:24.129
It can use your organization's data, internal documents, Gmail, anything on Drive.
00:14:24.129 --> 00:14:28.929
It can also connect to third-party data sources, so like your Salesforce.
00:14:28.929 --> 00:14:31.169
So that's something to consider.
00:14:31.649 --> 00:14:37.250
And then security controls, if you're using public Gemini, of course, there's no administrator.
00:14:37.250 --> 00:14:39.169
You're just out there in the public.
00:14:39.169 --> 00:14:41.169
So you have no control.
00:14:41.169 --> 00:14:54.610
For enterprise Gemini, it does have administrative controls, so IT can manage access, and that allows it to support data loss prevention and certain types of encryptions.
00:14:54.610 --> 00:15:08.529
So those are the quick uh updates on if you were confused about the differences between public and private AI tools for the two big players, Microsoft and Google.
00:15:08.529 --> 00:15:10.850
I just wanted to quickly go over it.
00:15:11.169 --> 00:15:28.529
If you're interested in other tools like Claude, Anthropics Claude, or Perplexity or Change AI, and you would want to just be aware, if you're creating a free account, you're essentially using the public version.
00:15:28.529 --> 00:15:39.250
If you purchase a subscription either for yourself through your workplace or for your workplace and enterprise version, then you have those additional protections.
00:15:39.250 --> 00:15:42.929
Your data is private to your organization.
00:15:42.929 --> 00:15:45.490
It's still super, super functional.
00:15:45.490 --> 00:15:47.649
So I want to emphasize that.
00:15:47.649 --> 00:16:06.210
It's just that, especially if you are concerned, you have sensitive information, you just want to protect your privacy and the privacy of your data, which of course everyone does, we highly strongly recommend to use the enterprise version of the AI tool that you're interested in.
00:16:07.169 --> 00:16:34.289
Of course, if you are using a tool that is updating you with AI, so for example, Zoom, uh Zapier, maybe your CRM, Salesforce, if they have AI and you like do the update and you now you have AI, that is also going to, for the most part, be controlled within your enterprise license for that tool for Zoom or Zapier or Salesforce.
00:16:34.289 --> 00:16:43.250
But you know, it's probably good practice to continue to look at the terms and conditions and just make sure that you're using AI safely at work.
00:16:43.250 --> 00:16:44.850
All right.
00:16:45.009 --> 00:16:49.649
Well, that's it for this episode, midweek check in on nonprofit AI.
00:16:49.649 --> 00:16:54.289
And I'll see you again on Friday with our regularly scheduled podcast.
00:16:54.289 --> 00:16:59.889
And again, every Tuesday we'll be coming back with some quick AI tips for nonprofits.
00:16:59.889 --> 00:17:00.769
Take care.
