Restrictions and ethics of AI: A must watch

- So this video right here is probably the most important one in this course, so do not skip this one. And I know the topic might sound a little boring, but it is really mission critical for all of us to understand what's going on with these chatbots, like, from an ethical perspective and to just also understand how to really treat the responses we're getting out of them and how to use these tools responsibly. I cannot have you go out of this course and start using these tools without knowing this section because it is that important. So taking a look here, we're just going to go over the restrictions of the AI chatbots, like what they can do and what they can't do, and we're going to get a little bit into the ethics. So taking a look here, which I would not recommend trying to do any of these more harmful things with the AI chatbots 'cause it actually might kick you out sometimes if you try to do things like that. It'll be like, "Hey, we noticed "you're going against our guidelines." But I did want to point out things that these AI chatbots are not programmed to do, which include providing legal or medical advice. And, I mean, you can ask it legal or medical questions. It will answer them, but it's one of those things where it is not professional advice, right? So we don't want to be treating it in that way. It also cannot perform criminal activities, so obviously do not want to be doing any of that to begin with, but especially not with these chatbots. It cannot disclose confidential information, so you can't ask it to like get somebody's bank account or something terrible like that. Don't want to be doing that. It won't use offensive language, which is cool. It will not create scams, it will not create phishing emails for you. It will not spread misinformation and it will not promote hate speech and it will not provide financial advice. Yes, you can ask it financial questions, but remember, similar to the legal advice or the medical advice, it is not actually like you are sitting down with a licensed professional getting that advice, right? Like, this is still just pulling from the internet and things that were used to train these different models, so it's really important to not treat it as, like, say a doctor and follow the advice, right? There is a big line that we have to draw here in terms of how we treat the responses that we're getting out of it. And now, this right here is the golden rule of all things AI that we need to remember, is that we must, must, must, underline the word "must" there, verify the AI chatbot responses and the sources. So it's really important to not just take the output that comes out of these chatbots and just send it out, right? You need to really be going through with a critical eye, analyzing the response, seeing what it came out with. And if you're using it to source information, make sure you're double checking those sources. Make sure that they're coming from credible places, and I'll show you in a sec how we can actually go in and do that, but it's really important to know that ChatGPT isn't just this like all-knowing thing or these AI chatbots, right? Like, it is something that pulls from the internet, and we know that anybody could really post anything they want on the internet, so when it's pulling those answers, they could be a little bit biased, they could be a little unethical, so it's important for us to really go in with our human selves, right? And take a look at that and give it our own touch and make sure that we're not just regurgitating information. And just like a funny story, I saw this on Instagram, where a student was going in and asking ChatGPT to write its paper, which I would not recommend doing. And it said, like, "As an AI chatbot, "I cannot write your paper for you, "but here is how I would." And it gave this information, and the student didn't even look. Literally just copy-pasted it with that top part in there and sent it to their teacher, and the teacher's like, "Well, you clearly used an AI chatbot to write this." Like, it's really important to, one, not use it for purposes like that, but two, really make sure you are reading and going through with a fine tooth comb all the output and not just fully trusting it and being like, "All right, ChatGPT said so, let's go." And really one of the reasons why is because these bots can hallucinate answers if they don't know them. And if you're like, "What?" That is the term that they typically use, is hallucinate. So for example, right now, as of this video, which you can always go in and check what the current cutoff date is, but the current cutoff date for ChatGPT is September of 2021. So as of the data that we used to train the model, anything that is past September 2021, instead of being like, "I don't know the answer," it might make up an answer, and if you don't know that, you might be like, "Aha, ChatGPT, thank you for the answer," right? And then you might use it and it might be completely false information, so that's why it's really important to also verify the sources. So when it comes to verifying sources from these different AI chatbots, these are some different prompts you see here on the screen that you can add into the prompts that you're writing, which we have a whole section on prompt engineering. Get hyped for that. But with this, these are some examples of different prompts you could use to verify, like, "Add links to credible sources," right? So be like, "Yo, tell me where you're getting "this information from." Include peer-reviewed scientific articles. That's another way to kind of make sure, like, all right, you're not just pulling from somebody's blog, right? You're pulling from some peer-reviewed scientific articles or reference, reputable sources, or provide with a table of sources just so you can kind of see, like, okay, where'd you pull that fact from, right? Like, we kind of want to know here, and one of the cool parts about Microsoft's Bing, which is one of my favorite ways to use it, is a lot of times when you're asking things, it'll give you the actual sources already on the bottom for you, so you can go in and be like, "Oh." And they're actually little clickable links, so you can click the link and double check the article and be like, "All right, this is a sound source "that I'm comfortable pulling from." So that's a really cool thing to keep in mind when you're going through these tools. We want to, one, always verify the information, always read the information, and don't just copy-paste it at face value. We always need to be double, triple checking these things.