Speaker 1: I love using ChatGPT, but many businesses who care about privacy including law firms, banks, and many tech companies don't trust ChatGPT with their secrets and many don't allow their employees to use ChatGPT in their work. Is it relatively secure or is it just a loud mouth chatterbot that can't be trusted with your secrets? We're going to explore that on today's video and stay to the end because I have some ideas on how to use ChatGPT safely. Hi, my name is John Ferrell. I'm a Silicon Valley intellectual property attorney. Welcome back to my channel. ChatGPT is an amazing productivity and writing tool. It's no wonder that as of this filming, OpenAI reports more than 100 million users and over a billion monthly visits. You type in a question or a prompt and it responds with near perfect grammar in more than 50 languages. And in many cases, its answers are so well written and its arguments are so perfectly formed that it's understandable why people want to use this as a productivity tool and to improve their efficiency. But before we look at the privacy issues, I think it'll be helpful to take a 30-second technical review on how ChatGPT actually works. ChatGPT is a natural language processing tool that uses machine learning to generate text responses from text prompts. It's based on a neural network, a type of algorithmic computing architecture that trains on a massive amount of language input data, such as articles and books and computer web pages. From this training, ChatGPT looks for patterns in text, words and phrases that commonly follow one after the other. And it's important to understand that ChatGPT and other AI models, at least currently, do not generate creative thought on their own. They simply create text based on logical patterns of words and phrases they've seen before. Now, all the power and capability of ChatGPT and other AI models raise concerns about security. For example, once inside your network, these AI tools would be certainly capable of generating malicious software code to attack your infrastructure or steal your data. And with their human sounding responses, these AI tools would certainly be capable of some pernicious social engineering or dangerous phishing attacks to gather passwords. And of course, the subject of this video, which is privacy and the collection of sensitive data and the possibility of misusing that data, either by the AI itself or those that collect and evaluate the data and use the data for training the AI models. Now turning to privacy, let's begin by looking at the privacy policy of OpenAI and remembering that OpenAI is the publisher or the company that owns ChatGPT. OpenAI principally gets information from three sources. It gets account information of its users. For paid users, this would include names and business names, addresses and credit card numbers. Secondly, OpenAI collects device information. This is information about your connection, the device type that you're using, the browser and your IP address. And finally, and probably most importantly, ChatGPT collects information about your text prompts and the responses that it gives you. Any interaction that you have with ChatGPT is recorded and stored by OpenAI. Now OpenAI's privacy policy is vague on who has access to the data that's collected and stored by ChatGPT. Now OpenAI states that it shares its data with vendors and service providers, with businesses, with affiliates. Now I'm not quite sure what affiliates means in this context, but since Microsoft is a significant owner in ChatGPT and OpenAI, we can assume that Microsoft has access to this data. The OpenAI privacy policy goes on to state that legal entities will have access to your data, as well as AI trainers who will be reviewing your conversations. Now it's clear that ChatGPT will be recording and storing the conversations you have, the prompts that you enter, the responses that you receive back. Any personal information that you include in your prompts will be remembered and potentially tied to your account information. Now it's one thing if I'm asking ChatGPT to give me some advice on mail pattern balding, but it's a completely separate problem if I'm asking ChatGPT to review an SEC filing document, for example, that I'm drafting on behalf of one of my public companies. Now it might be a little embarrassing to me if the world were to find out that I'm shopping for Rogaine, but it would be a completely different problem if an SEC filing document I'm drafting were to be prematurely disclosed. And the problem even compounds more when using third-party applications that connect to ChatGPT through their API. And we're going to see more and more of these third-party applications proliferate as time goes on. We need to pay special attention when applications that we're using say that they're powered by artificial intelligence. I would love to use ChatGPT and other AI tools to help me draft patent applications and contracts and to formulate litigation strategies. And there are a lot of other tedious legal tasks that I would love artificial intelligence to help me with. But at least right now, I and other lawyers can't risk the disclosure of client confidential information with these AI tools. And likewise in your work, you may have lots of tasks that you would love to kick over to ChatGPT to get help with. Reviewing and condensing meeting notes, or summarizing a company white paper into a list of bullet points, or possibly asking ChatGPT to help you debug a computer program that you're working on. There's so many ways that ChatGPT and other AI tools can help us at work, but we have to be careful about the risk of loss of trade secrets. Okay, as I said at the beginning of the video, I love using ChatGPT, but there are some privacy concerns. And so we have to be super careful about how we use this tool. And I have some suggestions on how you can at least have a chance of improving your privacy when using ChatGPT and other AI tools. So the first suggestion I have with respect to ChatGPT is to consider using the incognito mode. And you can find this by going to settings in ChatGPT. And under data controls, turn off the chat history and training. And presumably with this setting turned off, ChatGPT will not use your conversation data for model training and will forget your conversation or delete your conversation after 30 days. A second suggestion I have, and this was taught to me by my ninth grade high school civics teacher and has worked so well and has been so valuable for my clients. Never put anything in writing that you would be embarrassed to have read in open court. And I think that doubly applies for ChatGPT. This is not the place to enter prompts or questions about how to build a meth lab in your laundry room on both a personal level and in your business work. Until we get better clarity from OpenAI and other AI tools manufacturers on how they're going to protect our privacy, we just need to be very careful about how we use these tools. Now if you have a sensitive question you'd like to ask ChatGPT, consider using a throwaway email address. Now this might be a gmail address or some other email address that you've never used before and it would be somewhat difficult for ChatGPT to associate with you. Log into ChatGPT using a free account that you open with this throwaway email address and while you're at it, use the incognito mode on your computer browser to interact with ChatGPT. You want to make it as difficult as possible for ChatGPT to tie your identity to any questions or prompts that you might be entering. And the final piece of advice I have with respect to using ChatGPT in a private way is to be a listener and not a talker. When you ask ChatGPT a question, keep the questions general. Avoid pasting large blocks of text or memos that you've written or other large quantities of information into ChatGPT. Just assume that everything that you type into ChatGPT is going to give ChatGPT in the world more information about who you are. Okay that's all I've got on ChatGPT for today. If you found this video interesting, I think you'll find this next one helpful as well. I'll see you over there.