Beyond the Hype: Creativity with ChatGPT and Midjourney in under 10 Minutes

1. Introduction: AI the end of human creativity? I don't think it is, but we've all seen the souls posts on social media that are generated by AI and the shill AI. Say it's democratizing creativity. My name is Jasper Bam, and I have a PhD in machine learning, and I've been working in the field of machine learning and AI for over seven years. But I've been a content creator for 30. I'm a teacher here on Skillshare. I've taught over 10,000 people over the last years about machine learning, data science, and AI. I like to think about AI in non hype ways because I don't actually work on these language model. I predict the weather using AI. Is reflected in my content online. I'm a YouTube partner, and I write a newsletter that is read by over 1,000 people every week about non high AI, about the fundamentals and the real understanding of this technology. With this, I would like to invite you on a journey. T generative AI, using Chachi PE and mid Journey, having a look at how they work, but also the limitations and realistic use cases and explore how we can co create with AI without losing that human touch. 2. Class Project: To the course. Glad you chose to go on this journey with me. In this project, I would like to keep it super focused because it's a short course, so it shouldn't be a huge project. I would like you to create two pieces of content to compare and contrast, take something that you really know something about. This can be desperate housewives, or this can be geology, take a simple prompt, and then use your expertise to actually refine this. So have a conversation with the AI to either make a better image or make a better text. This is up to you. I'm open to both, and I'm really interested to see what you do. 3. Capabilities of Generative AI: AI and machine learning models are models that learn from data. With text AI, we are talking about large language models. These models basically take a bunch of texts that is stolen from the Internet and learn how to predict the next word in a sequence. So when you prompted, tell me a bedtime story, it basically has a statistical model, so a mathematical model. So then it goes word by word and generates the story. But in the end, it's just fancy text prediction. Image models work a little bit different. We basically create noise, so just random numbers. And from that, we are using the prompt that we use to, well, maybe get a cute cat picture. And this is slowly converting that noise into an image. Basically chiseling away from this noise to, well, in the end, get a cute cat. What you'll notice is, they don't actually think they are statistical models that predict text or that predict how to get rid of noise. They don't actually think, Oh, I know what a cat looks like. 4. Limitations and Shortcomings of Generative AI: Look at the limitations and shortcomings of generative AI. The first one is the lack of true understanding. We have to be really careful. GPT sometimes can seem like it understands language. But a famous example from social media is asking it, how many are in the word strawberry. There are three, but currently, all of the LLMs actually say there's two. It can't actually represent its own language output. Another problem with Chachi PT and M Journey is the lack of originality and novelty. Now, these models are trained on all the text from the Internet. But we can actually see that when people create new stories, they are more along the lines of things that already exist. The same with images. These images are always perfectly in focus, and the subject is right here in the middle of the frame. It's actually kind of difficult to get to produce bad images. And we can even see this lack of novelty and reuse of certain words. We see this in text analysis of papers submitted to AI conferences and words like Delve, unleash, unlock rated highly on the confidence scale. But Delv is also something that we use a lot outside of the US. English is my second language, and I've been using Delvinto for a really long time. But now, well, I sound like I use J GPT. The third drawback is genuine human connection and well, emotional capability. Our emotions and our internal monologue never really gets published. So it is not in the training data. Our deep internal motivations will never be captured. And I view that when we read these stories because these characters are sometimes a little bit too dimensional. A fourth and final thing to consider our ethical concerns. I talked about these models basically being trained on stolen information. The problem that Chachi PD, for example, has outsourced some of its content moderation to Kenyan workers for $2. Mid journey, and for all these image generators, we are looking for high rise images. We naturally find an under representation in the training data, something to keep in mind that there is also a bias, what is even represented in the learning of these models. 5. Where AI Truly Adds Value: Let's look where Ctrip and Majury can actually add value because these models and these tools can be a great ally in our creative process. For me, one of the biggest ones and one that I regularly use is brainstorming and outlining, because those don't rely so much on generating text and making sure that text is correct. But those give structures and outlines, and those are also easy for me to verify and easy for me to edit and add to. Usually, in Chachi P, what you can do is have your story and generate different plot twists in there. But yeah, basically give you the idea of what's going to happen. For mid Journey, I would use it very similarly to generate different ideas to get, like, different visual identities for like a branding project, different perspectives on a certain art project, compositions of images, and just have an overview of what you like more, and then actually create it yourself. Other really nice way to use Cat PT and M journey as your ally in the creative process to streamline repetitive tasks. We all have to write e mails, and I actually use a tool for neurodivergent people called Goblin Tools. It's free. Check it out. And yeah, it's fantastic to get your e mails at the right tone. The third one is to create data driven insights. So these machine learning models are extremely good at pattern matching. So basically, if you put down your current interests and want to combine them into a certain format to be able to generate outlines generate text for you very quickly to get an idea of a certain complex task. And finally, it can be fantastic to co create and enhance collaboration. It's not always possible to work with other artists on a project. So when you're writing or when you're creating images, it can be really nice to get feedback and go into a rapid iteration cycle. With AI that isn't always the best at giving you novel ideas, but it can give you feedback on something that you've written and basically give you new well, avenues that you could take this next or well, critique some of your writing and really get this to upgrade your creativity in this. 6. Finding the Balance: Human-AI Collaboration: AI for me, really is about gaining a new ally in the creative process. It's about co creation and collaboration. And there's a really interesting study about how our writing changes by using Cha chiPT. They have three different types of people, one that didn't use AI at all, one that was able to use one prompt. And one that was able to use up to five prongs. And they found out, which is fascinating that the writing after this was actually judged as more professional, more interesting, and actually as a better story, and the more AI they used, the better the stories were judged by their peers. However, those stories were then much more similar to each other. It's kind of funny in a way, right? We're now arguably writing a better story. But you're lost in a sea of the same goodness. We're losing a little bit of our individualism. That's why I really like to edit content and make it my own. The important part about content creation and creativity isn't a novel story. We've all created the same story over and over again. We know the hero's journey, we know the three acts, but we have to do the work to elevate it out of a sea of nothingness. 7. What We Learned: That's it. If you have any more questions, leave them in the comments. We learned about the ethical implications, the drawbacks, but also the fantastic real life use cases. Thank you so much for tuning in. I hope to see you in another course.