What is ChatGPT? Understanding the New Face of AI
Discover the power and potential of Generative AI and Large Language Models. Check out this episode to learn about these amazing/terrifying technologies.
10 Easy Ways to Immediately Boost Your Online Security – Resilience Cybersecurity & Data Privacy
8 Useful Small Business Cybersecurity Tips You Need to Know – Resilience Cybersecurity & Data Privacy
What Is Generative AI? – McKinsey & Company
What Is Artificial General Intelligence – TechTarget
Artificial intelligence (AI) vs. machine learning (ML) – Google Cloud
Calm Down. There is No Conscious A.I. – Gizmodo
How to Use AI in Cybersecurity and Avoid Being Trapped – The Hacker News
ChatGPT Used to Develop New Malicious Tools – infoSecurity Group
Brian: Hey everybody. Welcome to Fearless Paranoia. Thank you for joining us here where we seek to demystify the complex and confusing world of cybersecurity. I’m Brian, the cybersecurity attorney joined by Ryan, the cybersecurity architect. Thank you all for enduring my absence for the last two weeks, I’m sure it was rough courtesy of a case of pneumonia given to me by my son, I could pretty much barely talk for the last two weeks, my wife thoroughly enjoyed the silence, but my voice has returned. So I am back here to talk to all of you. So we’re going to have an interesting series of episodes here, we don’t normally do continuous multi-part episodes like this, usually we’ll do one subset of a particular topic. But there is an issue that has come up lately that we really want to talk about. And there’s just no way to talk about it succinctly in one episode and keep it digestible as we want these episodes to be. So what we’re going to be talking about for the next three episodes, we’ll just call it the What is ChatGPT discussion, we’re going to be talking about artificial intelligence, and we’re going to talk about it in three parts. This episode is going to be discussing primarily the definition behind what we actually mean we’re talking about AI one of the most frustrating things for me over the past few months in reading about all of these new technologies, ChatGPT, Bard, all the generative AI image development stuff is that everyone talks about AI and no one seems to have any really comprehensive idea of what they’re talking about. There’s the Sci-Fi version of AI, which is really what a lot of people think of, which is the autonomous machine that learns, operates independently, everything like that, and actually uses human-level intellect. That’s what most analysis describes as general AI. Well, general AI does not exist right now. And most real experts on general AI believe that it is 50 to 60. To Infinity years away. general AI is a concept that a lot of very, very smart, very serious people believe may never happen. It is a complex thing that our brains do on a second-by-second basis. Okay, so we’re going to talk about what AI means. And we’re gonna give you some examples of a few of the things that currently exist. When we talk about AI that what we’re talking about the second episode, we’re going to talk about how we got here and where we’re going, what is the whole drive behind AI? What benefits do we think we’re gonna get out of it? And what are some of the key things that both Ryan and I kind of anticipate as being great benefits of some of this technology moving forward? The third episode, we’re gonna talk about the scary stuff. This is what we fear, and what we’ve heard others describe as their fears of this AI, we’ll call it AI menace, everything from the existential to the banal and boring, yet still phenomenally dangerous, potential downsides to the use of the AI tools that are out there right now. So with that, we’re gonna get started. Ryan, is there anything else you wanted to mention about what we’re gonna be talking about here? Before we dive into these definitions?
Ryan: Nah, I think we can just jump in and start really digging into these is going to be a really fun discussion. And I think there’s a lot to cover.
Brian: All right, well, right off the bat, what we’re going to do is we’re going to talk about the definition of what AI is artificial intelligence in a broad context is the enabling of a machine or a system to operate like a human to stands to reason to act to adapt. And it is a combination of various tools and devices, a whole number of different aspects of the ways a machine can operate and react and act more like a human. Then there’s the more specific topic of generative AI. Now we’ve talked about general AI just second ago, general AI is the concept of how you know, the self-sufficient, self-operating computer system that thinks on its own generative AI is actually probably the big thing we’re dealing with now. And generative AI is a type of AI that produces stuff, it writes things, it creates pictures, imagery, audio, video, that’s really the big subject of what we’ve been dealing with lately is either text-based or generative AI. And I think the most specific thing we can talk about there is the concept of generating a response to human interaction, we are actually going to discuss machine learning in a future episode. It’s a very interesting topic. But people very frequently confused machine learning and AI. Machine learning is a component of an artificial intelligence system. Machine learning is when a machine is able to learn it’s when you set up a system to have specific inputs to create a response based on those inputs that improve over time and that it learns from its own answers based on an algorithm. Machine learning, however, is only a part of AI and machine learning actually does itself exist entirely separately from AI as well. Machine learning does not require AI to operate. We’re not going to be talking about machine learning specifically indirectly but just know that it is included in what we’re talking about. Okay, right. I’ve just said a lot of words. I’m good at saying those words. Can you do me a favor and just tell me what those words together meant?
Ryan: It was a lot of words, what it really means is that as human beings, we’re looking for ways to continue to branch out and use these technological tools that were developed to further automate tasks ever since we put together things like the assembly line and other automation, that’s just been human nature is to find ways to take those general repeatable, rather boring, but necessary tasks and find ways to automate them. And in this case, this is trying to take the next step of evolution into quite a few other spaces by jumping into this type of use of computing power to generate similar output to what we would produce naturally as humans, rather than just producing an output based on computational inputs has been kind of the nature or has been the nature of computing for quite a few decades now. So this is really just kind of a next step where we are starting to step away from just using computers to do computing things. And now we’re looking at the opportunity to use computers to do human things, which is relatively new. And so jumping into things like generative AI, and all of the underlying things like large language models, which make up the power and the capacity behind it is really, really a first interesting step. And we’re starting to see it be used for a whole variety of things. And it’s being adopted by a variety of different companies, it’s gotten a lot of traction. And it’s really going to be interesting to see how many different good use cases we can find as people to take advantage of these advances in technology.
Brian: So you said something right there and want to try to do as many of the definitions of help people understand what we’re talking about. So anyone who’s heard of ChatGPT over the last six months, I mean, I guess it really didn’t launch until the end of November. But by January, it had something like 100 million users, it was the fastest-growing app. There’s a poll I saw recently that said, half of legal professionals responded to a survey saying that they use ChatGPT or a similar large language model as part of their daily or they use it on a daily basis. And I immediately question and said, Okay, this survey, how are they defining legal professionals? How are they defining us? And how are they defining daily because the bottom line is, I don’t know that many people who have yet found a way to incorporate ChatGPT or a similar system into their daily routines, that’s it hasn’t happened yet, it will happen just the same way a computer became a part of the daily routine, that sort of thing will likely happen, because it makes a lot of things faster. But at the same time, people talk about What is ChatGPT and talked about it as AI but it is a very specific type of AI. And you mentioned the term large language model. Now I know large language model is actually when I mentioned machine learning earlier, the whole point of the large language model is it’s a type of machine learning algorithm. That’s how it functions, I guess, talk about what large language model means. What do you mean when you say that?
Ryan: Oh, we can talk about two different ways. Let’s take a real quick and let’s let the large language model and the generative AI tell you really quick what the differences large language models as its use of artificial neural networks with a large number of parameters that have been trained on massive amounts of text data. These models are capable of generating coherent and fluent text that mimics human language primarily used for language generation tasks, such as text completion, language translation, text summarization, generative AI, on the other hand, refers to a broader category of artificial intelligence capable of generating new content or outputs, such as images, videos, music based on a set of rules or inputs. This can include techniques such as deep learning, reinforcement learning and evolutionary algorithms. So that’s the definition from ChatGPT right there of what it is that makes up what is ChatGPT. And but if we want to really simplify it down at an input and processing and an output, and really at its core, that’s how these two are kind of interacting with one another. And that’s why they have a lot of overlap because one is being used to bring in large sets of data and do interesting things with it and prepare a lot of interesting capabilities with it. The other tool is reaching into those capabilities and finding ways to generate creative and interesting outputs that people actually want to see using the models that are put together.
Brian: You’re listening to the Fearless Paranoia podcast. For more information on keeping yourself your family and your company protected against cyber threats, check out the Resilience Cybersecurity and Data Privacy blog. If you’re enjoying this podcast, please like and subscribe using any of your favorite podcast platforms. Also, please share this podcast with anyone you think would find it helpful or useful. We rely on listeners like you to help get the word out about this show, and we appreciate the support. Now, time for some more cybersecurity…
Brian: If we were to talk about just on a step-by-step basis than what it is you got your input, which is as it says text and data. Well, we know ChatGPT basically and we’ll get to this later essentially crawled the internet and read text and it was fed in and over years and we’re talking about years here that input was categorized by human reviewers and listed and values were placed onto it because with machine learning there has to be value, it can’t just be an unorganized cat. So you have values so that the machine knows what’s being put in. And then based on an algorithm, it takes those inputs and creates an output. Now the next step on there, the generative AI portion is that output then has rules applied to it, and essentially said, Okay, so we’re going to ask based on its inputs, what the output should be. And then we’re going to take the output and configure it in a way that it’s understandable language, you know, it might have a response that has a ton of nouns and verbs in a random order. But that’s simply not how things work. So you have rules that further narrow it down into something that’s understandable, and also responsive to the question. And also probably not, you know, providing for example, you know, details from the Anarchist Cookbook and other things like that, you know, rules that you content rules. So you make sure that you’re not allowing dangerous people to require dangerous knowledge simply and easily over the internet.
Ryan: Yeah, especially the very last part about the input sanitization. That’s a huge piece in cybersecurity that everybody has a job everybody in cybersecurity wants to beat, I think is, is that one, but yeah, making sure you’ve got great data to use and great processing, and finding ways to derive great about. I mean, you had all the nails right on there.
Brian: Okay. So that’s, I think, a good point to shift away from what we mean, what we’re saying to what we’re actually talking about, because the bottom line here, there’s so many different things we can talk about, of what this stuff does, and various options. So we’ll start with text. Now, we know that there are text models, and among those is what is ChatGPT. But there’s also discussion about what is ChatGPT being a chatbot, what does that mean? What is the difference between those?
Ryan: Yeah, I mean, realistically, again, it’s a chatbot. So it’s made to take human interaction, and it’s made to provide an equivalent interaction back. And so that’s really the underlying core of that type of interactive chat implementation of generative AI, it needs to provide not just structured output, but one that is conversational in nature. So realistically, the outputs are fine-tuned around recorded conversation or documented conversation to kind of fall in line with the same grammatical uses and the same sentence structure that we would otherwise use rather than just deriving a route. But other AI obviously, would tailor its outputs based on the specific use cases, and whatever is the most effective way to provide those outputs. And in this case, GPT provides something that’s hopefully easy to read. Makes sense, which it doesn’t always but that’s the point behind it. And you know, it’ll get there. But that’s the point is it’s made to be conversational, effectively.
Brian: And like that same thing would be true of barred and bing chat is always listed as similar but these are chatbots designed to interact with you and receive an answer me basically, it’s if you turn Ask Jeeves into an actual response…
Ryan: It takes it one step further, though with a generative AI. So with something like Ask Jeeves or Google, you can easily format in an entire question. I know people that will just pick up their phone and just say, whatever, Hey, Siri, or hey, Google, whatever. And we’ll ask it an entire question. All of that just gets pasted right into the search bar, basically. And then it does its normal search engine, you know, rating of whatever, and it produces an output for you. In this case, it’s not just producing outputs, it’s scary, it will produce outputs. But it’s not just producing a list of outputs. It’s producing conversational return output back to you. But then the generative AI models is also made to take that one step further. Now, if you go and say, get a set of search results from Google, and then you go ask Google, hey, what would you think of a way to parse through the search results? Or what’s your opinion on a piece or some other aspect of the search results, Google is going to treat that as an entirely new search. And it’s going to go through and produce a whole new set of results, it’s not going to relate that in any fashion directly to your previous search. The conversational aspect of something like ChatGPT allows you to take those previous outputs that it’s even produced for you and then take another look at them from another aspect. I’ve had a couple conversations with ChatGPT, where I said, I’m not happy with this output, can we look at this from a different perspective, and all of a sudden it will reassess everything that is given to you based on those new criteria and provide a new set of outputs. And so it’s a different level of being able to kind of do that communication. So if something like that can be applied to search engines in the future, it’s going to make the ability to get to the data you want much more efficient because now you can just ask a few questions, continually tweak your search until you get right to the piece that you need. Right now. We’re just kind of punching it into Google maybe after redo the Google Search four or five, six times to get exactly the keywords that are the right ones or you just need to go through page after page after page to look for what you’re trying to find here. You can help use this tool and continually just through conversation guide that path closer to its finish line much quicker and the tool will just help you get there.
Brian: You’re listening to the Fearless Paranoia podcast, we’re here to help make the complex language of cybersecurity understandable. So if there are topics or issues that you’d like Ryan and I to break down in an episode, send us an email at firstname.lastname@example.org or reach out to us on Facebook or LinkedIn. For more information about today’s episode, be sure to check out Fearless Paranoia.com where you’ll find a full transcript as well as links to helpful resources and any research and reports discussed during this episode. While you’re there, check out our other posts and podcasts as well as additional helpful resources for learning about cybersecurity. Now, back to the show.
Brian: And then, of course, there are all sorts of other tools. When we’re talking about generative AI. I do include these chatbots as generative AI because even though the primary driver behind them and the things everyone’s talking about is the large language models, they are taking the output of the large language model putting it into a conversational form, which means they’re creating content, and they’re creating it responsive to something and the interactive nature of it in my mind even creates an even greater reason to define it as generative AI and some of the other big ones that have been making the news lately, there are plenty of image AIs, I think that probably the Stable Diffusion and Mid Journey probably the most talked about recently, I’m blanking right now on which one of them got funded by Getty Images was because they, they kept generating they stole so many pictures that were watermark for getting image that a bunch of their generated images actually had the Getty Images watermark on it, to me…
Ryan: Mid Journey. A shot they did to the Pope was pretty phenomenal to the Pope and the trenchcoat, or whenever that was the big coffee, the big one puffy coat pocket. That was very, very entertaining. I use mid journey all the time, just for nothing more than personal enjoyment and entertainment. But yeah,
Brian: and that’s of course, now it’s I if I’m not mistaken, mid journey did just take down their basic free-to-use tier because of those post pictures. But it is one of those great examples of where people need to put that filter on, when they sit in front of the computer that is critical. You really need to channel Socrates there and question everything and just be able to and prepared to separate everything you see as not necessarily real life, because of the things that can be created. I mean, there are plenty of email response once it’ll write an email response based on the input and based on your previous emails that you plug in. There are the GitHub copilot is a huge one I know is being used a lot for people who want to develop code. I question the sincerity of anyone who says that helping people write malicious code is one of the scariest things about this because I tend to think that the people who are willing to rely on at least the current versions of generative AI to create malicious code, we’re going to do it anyways. But you are to take some of the easier steps out of the process.
Ryan: Most of the malicious code so far that people have tried to produce coming out of these engines tends to be quite poorly coded, quickly picked up and fingerprinted. And I don’t think that I think it’s going to be a while and it’s going to take someone that’s got really creative ability to speak the GPT language basically, to know how to ask the right things to get the code tailored the way it needs to be before they’re ever going to be really effective at that. So I think you’re gonna see the same way you had people learn Google foo back in the day and got to be your your really big Google experts, I think people are going to have to figure out whatever it’s going to be LLM, foo, generative AI food, whatever the next food is going to be, that’s going to be the thing that’s going to take us forward. But I think that in order to really get that effective of an output, at least anytime in the near future, while this is still under development is going to take some really interesting creativity. And most of the developers of this are pretty well aware of it. And those putting large amounts of money into it are actively working to combat this because they understand they don’t want their projects to be tarnished by having this kind of abuse to it, especially when it’s in these early stages. And they’re still trying to get their foothold here.
Brian: There’s just a lot of tools out there that are using generative AI systems to face if this were the most likely outcome of any development on this front is not going to be a mass replacement. But with mass augmentation, we’re not all going to be replaced by robots, we’re all going to become cyborgs. That’s the takeaway that I’ve got from this as a person who has spent a lot of time over the last 10 years as a content creator, I look at a lot of these and I have fears, but I also have that feeling you get in your stomach where it’s still you’re a little bit more excited about the potential promise. What is ChatGPT? And it’s a very interesting subject. And it’s one we’re going to be getting to shortly we are out of time for this particular episode. But make sure you tune in next week and the week after next we have the remaining two parts of this three-part series where we discussed the promise of generative AI and then the peril of generative AI I want to thank you for joining us today remind everybody that you can subscribe to Fearless Paranoia through your favorite podcasting apps or through our website or any of our social media sites. Please please do share this episode out if you find it useful or know anyone who could benefit from actually understanding what they’re talking about when they talk about AI. I think it’s a pretty valuable little tool, but we’re going to really enjoy discussing the rest of this with you on behalf of Fearless Paranoia. I am Brian
Ryan: And I am Ryan
Brian: And we’ll see you next time.
to make cybersecurity understandable, digestable, and guide you through being able to understand what you and your business need to focus on in order to get the most benefit for your cybersecurity spend.
©2022 Fearless Paranoia