S1. Ep24 - The Hidden Privacy Settings in ChatGPT That Could Expose Your IP

Katie: Welcome to the AI Automations for Business Podcasts. We're your hosts, Katie and Null, and we'll be discussing how you can use AI for your business along with the latest news, updates and automations to help you stay ahead of the curve, allowing you to grow and scale your business more efficiently. Please be sure to subscribe and I hope you enjoy this episode. Welcome back to another episode. Hello, I'm Katie. And as always, I've got Noel here with me. Hey Noel.

Noel: Hello.

Katie: So we were having a conversation yesterday, which has sparked the reason for this podcast episode, right? Yes. This thought entered my head because I had seen a, I think like a Facebook post somewhere. And someone had said, like, please don't upload my slides into chat. And I was thinking, that's interesting because actually once that's out in the public, you don't really have control over that sort of thing, do we? Like, we never really have.

Noel: No. I know, I guess also it kind of like gives you the idea to actually input it. If some people might read that and think, oh, I didn't think about putting it in there and then put it in there anyway. Yeah. Yeah.

Katie: So I was thinking, oh, I wonder why this person doesn't want those, you know, slides uploaded. So I kind of like scrolled back and like clicked on it. And I kind of hadn't really thought about what does happen when other people upload your content into chat GPT or LLM. And what does that mean? do, like, other people then have access to, like, your paid things and things like that? So this is where we were having this conversation, wasn't it?

Noel: Yeah, yeah. So deep, dark hole to jump into.

Katie: Yeah, this is what goes on in my brain. So we were having this conversation. So I said to know if I had attended, like, your paid, let's say, webinar master class. And at the end of the webinar masterclass, you said to everyone, I'm going to give everyone my slides. So I've attended, I've paid, and then I've got your slides, and then I'm going to upload them into chat GPT. What does that mean? Like, does it then mean it's like, then, like, in the AI world and can then anyone else, like, essentially then use your content to help them create their content. Yeah.

Noel: Yeah, it's kind of a grey area because it can and it can't at the same time. So yeah, it's kind of important to understand the difference between the two, but then you don't really have any control over someone else's chat GPT settings. So it is very tricky to kind of say, well, you know, you need to tick this box, But if they don't tick a box, then, you know, it could be available and being used to train the next version of chat GPT, which isn't ideal if you want to keep that secret.

Katie: Yeah. Because you were saying, as well, yesterday, like, so we've obviously now gone and done a bit more research on this today after this conversation that spiralled into, yeah, a deep black hole. But originally we thought that only on the free Chachupit account could you not turn off, like, don't use my content to train the AI model? Yes.

Noel: So I think, yeah, because we've had Chachapit accounts almost since the very start. So that used to be the thing where it would be like, yeah, if you're free, we're always going to use your data. If you're paid, then you can turn it off. but yeah having a look now at their privacy on the pricing page I was a bit surprised to see the differences so now if you've got a free pro or a plus account so they basically say you can turn it off but by default the button is turned on so we will use your data to train our next version of like GPT5 for example or six whatever whatever silly name they're going to give it next. But yeah, but by default, that's always on. So it's up to you to go into the settings and it's under data controls. And then it just says, you know, improve models for everyone. And you can just tick it on or off. So that's, yeah. So that's kind of interesting. And I didn't realize that by default on the two, especially the $200 plan. I wasn't expecting that to be on by default. But it is.

Katie: So what does that actually mean then? So say you weren't aware of this because I don't think like a lot of people would be aware of this. You know, so what does it mean if you haven't had this like toggled off and, you know, AI has been using your content to train next models?

Noel: I guess if you're using it and having like your own IP, so like you say, where you've created a course, it's absolutely brilliant. You know, you've had AI's help you do it along the way, and that box isn't ticked, you know, so the data is being used to train AI. Then there's a potential that somebody else could ask to create a course based on the same topic, and it could come out with some very similar themes to yours.

Katie: Yeah.

Noel: And yeah, they get down the copyright dilemma, which we discussed with Lucy, didn't we, a few weeks ago? So it's like, who owns it?

Katie: Yeah, I was just like to say this then comes into what we were discussing with Lucy Wheeler on our AI legal episode, which if you haven't listened to, honestly, I would highly recommend you going and listening to that episode. We will link it below in the show notes for you. We'll make it nice and easy for you. But that was a really fascinating conversation and really helps you understand where you stand if something like that happens because this is the thing. Like people are uploading content and creating course ideals from ChatGBT. And then someone else is saying, you know, I want you to help me outline the course for this subject. And, you know, you're putting in, you know, a really. really great prompt you know you're being really specific about it but it then you know throws you out this um course outline you think great and you create your course and then you put it out you sell it and then only for like someone else to go hey you're actually copying my course and you're like uh no i'm not um so yeah we explained quite a bit of that with lucy on that podcast episode and she also explains about it more in her AI legal toolkit which again I feel like is an absolute must for any business owner using um you know chat chip tea or claude any AI at all we learned so much and we we also um thought about things as well that actually we hadn't thought about before.

Noel: Yeah.

Katie: It opened our eyes quite a lot, didn't it? Yeah.

Noel: I mean, we always did best practices, but yeah, even so there was still a few bits. Like, ah, right, I hadn't fully thought about that, you know, in that much detail. I was like, okay, yeah, we need to figure this out.

Katie: Yeah.

Noel: Yeah.

Katie: Okay. So now we are all obviously going to go and turn that setting off.

Noel: Especially, yeah, if you're putting any sort of business information, one of the interesting thing they do say is if you input personal information, they will do their best to ensure that that doesn't make it into the model training. But by saying it does its best, doesn't mean that it's 100%.

Katie: No.

Noel: So if you're putting any sort of client information through there, you probably shouldn't anyway, but, you know, if you do, there's potential that could also get leaked into the wider AI model world, which is, yeah, not ideal.

Katie: No, okay. And then there was something else you found out, you know, whilst using ChatGPT with the thumbs up and thumbs down.

Noel: Yes.

Katie: So obviously, when you're having a conversation with ChatGPT, at the end of like the conversation or whatever it like spits out at you at the bottom of that conversation there's always a thumbs up a thumbs down and then like two little arrows isn't there um tell us more about the thumbs up thumbs down

Noel: yeah i mean this is something they've never used i was surprised when i asked you yeah i use it all the time i was like why this somebody actually uses it

Katie: i like to give feedback.

Noel: I like to

Katie: tell when it's done a great job, I'm like, yeah, do what? That was really good. Or when it's really frustrating me, I'm like, no, you're not listening to me.

Noel: Well, there is an issue with doing that, which I didn't know myself, to be honest. So let's say you've turned off the ability for AI to be trained on your information. That's fine. That will work. But then as soon as you hit the third, thumbs up or thumbs down, that then gives Open AI the right to basically grab your entire conversation history and then use that to train this next model. So even if you've got it turned off and you, you know, say, yeah, that's good. It's going to grab all of that, put it back into the system and then use that as what, like, this was a good response for this question. Yeah. And then use that information later on. So yeah, that's kind of of important to understand.

Katie: Yeah. So if you do want to toggle off the option not to have AI to, you know, use your conversations to then train its next model. Also, you might really want to consider whether you use then the thumbs up and thumbs down at the end of those conversations as well, just something to think about.

Noel: Yes, definitely trained. Like if you've got a larger team. It's obviously then training your people that use it on a daily basis just to make sure that they know, you know, if you do tick these options or you, you know, do a thumbs up, thumbs down, then your business information is also going out the door.

Katie: This isn't ideal. And that's something else. Like I don't think a lot of people who have got teams really have actually got best practices or compliances for other team members or, you know, employees to use it. And I think actually this is going to be so, so important. If any of your employees or team members are using like any LLMs, then I just think it's so important to have those best practices and compliances. What, you know, you don't put in and what you do put in, you know, especially if it's sensitive information, naming, clients, yeah, that's definitely worth a conversation.

Noel: Yeah, but I guess if you have a bigger team, you're probably going to move to the team's plan anyway, or maybe even enterprise if you're a really big business. But just to make a note here as well is for those for teams and enterprise that option of sharing data is automatically turned off it's up to you to turn it back on so your default is off which is real handy

Katie: yeah really handy so what about when we're using AI automations and we're you know asking the AI to I don't know like create report or you know to scrape data and stuff is our information then like i guess i don't really want to say safe but is it kept private it's not being trained yes

Noel: that it's not using that data at all so yeah if we're using it well we're in an automation we'll be using the API to access it programmatically so if you're using that then they do not train any of their models based on anything that goes in or out using that API. And that's the same for, I know it's definitely the same for OpenAI and Anthropic. I believe Gemini is also the same. I think, yeah, Gemini is sticking out in my mind as maybe one that isn't, but I think that's GDPR. Anyway, but yeah, most of them, they don't do it, which is really good. So it's all safe. It's really well encrypted. And, yeah, they only keep the data for like 30 days. They don't access it. And then what's that 30 days is up is gone, gone forever.

Katie: Yeah.

Noel: So, yeah, very secure. There's no real need to worry. That's something that people always ask us and say, well, you know, because they know, like, chat, JBT, their stuff can be used to train the model. So they do get a bit, you know, worried and quite rightly worried. They're thinking, well, if I'm creating these documents and reports, which is like our IP, our special thing that we do, you know, I don't want that going somewhere else. And yeah, with the API, we're all good.

Katie: I mean, that's pretty nice. Yeah,

Noel: absolutely. I mean, if ever one of these guys was fined out for not doing that, then they'd probably go out of business very quickly.

Katie: Yeah.

Noel: So, yeah, yeah. It's very safe.

Katie: Okay. Well, I'm glad that we've had this conversation because I feel like for so many people it's going to be something maybe that they haven't even thought about.

Noel: So, yeah. I guess there's also one added extra bonus as well. So we had this conversation last night and while I was having my morning cup of tea, something popped up on YouTube and it was the new offline version of chat GBT. So Open AI released a new model, which you can download and have on your local computer and basically have ChatGBTGT fully offline. So that came out yesterday. So I've tested that this morning. It works well. Do you need a good computer, though? I tried run it on my MacBook M3 and it was like, nope.

Katie: Why wouldn't it run on your MacBook 3?

Noel: So it comes down to the GPU V-RAM to get technical about it. I can you explain

Katie: that for someone who isn't technical?

Noel: The V-RAM is what's kind of used to do all this complicated computing power. So if you have like a PC with a graphics card, that has all that V-Ram in there. So my PC has that and we have like 24 gigabytes or whatever in there. and these models, or the lowest model, needs at least 16 gigabytes of V-RAM. So that would work fine on my machine. But with a laptop, especially Mac, they tend not to have like graphics cards in there and that sort of computing power. So yeah, it fell over very quickly. My screen pretty much froze. I was trying to ask a simple question. So, yeah, that's not usable. But if you do have a higher-end PC, and things that will access to a graphics card, then, yeah, it's really, really good.

Katie: So it's unlikely it's going to work on a phone as well?

Noel: Oh, no, no, definitely not. Not yet anyway, no. Okay.

Katie: Maybe for the next updates.

Noel: Yeah, yeah, because what they do with these AI models is they train them on a set number of parameters. So they've released two, and they've got a really powerful one, which is based, on 120 billion parameters and then there's a smaller one which only does 20 billion so my graphics card's pretty decent and that could only do the 20 billion so yeah you know yeah you need real real superpower graphics cards to be doing the bigger version but the smaller ones just as good but that option's there and yeah

Katie: okay the options there yeah amazing okay well thank you so much for listening to this week's podcast episode. We're really interesting to know, like, have you ever thought of, you know, what happens to your content when you upload it to, say, chat GPT? Have you ever thought that, you know, it could be training the AI model based on what you are uploading onto it? Have you set best practices in your business or with your team and employees. Let's have conversation because I feel like this is so, so interesting. And again, we'll leave all the links that you need below and we will catch you next time for another episode very soon.