Episode 1: “Hello World + A Look at ChatGPT and Midjourney”
Listen to this episode:
Episode description:
In this episode I explain what the heck I am trying to achieve with this podcast, which is to provide introductory information about generative AI topics in short and sweet episodes. I don’t go too deep, but I plan to give enough information and tactics so that you can get started on your own. This episode covers ChatGPT and Midjourney, two of the most popular generative AI tools right now. ChatGPT generates text output and Midjourney creates image output. I run down how easy it is to start playing with ChatGPT and give examples of some things you can ask it to do, then dive into how to set up and start using Midjourney to create images.
ChatGPT
https://chat.openai.com
“45 ChatGPT Use Cases for Product Managers” by Aatir Abdul Rauf
“BuzzFeed says it will use AI to help create content, stock jumps 150%” at CNN
https://www.cnn.com/2023/01/26/media/buzzfeed-ai-content-creation
“ChatGPT is everything you wanted Bitcoin to be” by Josh Brown
https://thereformedbroker.com/2023/01/25/chatgpt-is-everything-you-wanted-bitcoin-to-be/
Midjourney
https://www.midjourney.com
Midjourney Quick Start Guide
https://docs.midjourney.com/docs/quick-start
DALL-E
https://labs.openai.com
Getting Started With Stable Diffusion: A Guide For Creators
Lore Machine Chief Prompt Officer Job Listing
https://www.loremachine.ai/cpo
Episode script:
Okay everyone, this is the first episode, so it’s gonna be a little raw. They should get better as I do more. But even if the audio isn’t incredible, or you don’t like the sound of my voice, or whatever.. My intent is to fill about 15 minutes or so with enough value that it doesn’t even matter. My name is Scot and I’ll be sharing what I’m learning about generative artificial intelligence. I’m taking a deep plunge into these revolutionary new tools, and trying to distill down the practical information for others. I hope you find it helpful.
Look, there are plenty of extensive YouTube videos, articles, and how-to guides on generative AI right now – so many that you may be overwhelmed. It takes time to sort through and consume all of that content. So my pitch is that I can give you plenty of nudges and tactics while you are on the go, or working on something else in your garage, or however you listen to podcasts. You can use the information later, like a quick start guide. I’ll include as many links as I can relative to what I am talking about in the episode description.
Okay, so there are A LOT of tools already available, right now in early 2023, and I hope to highlight my experiences with a different one, or several, or a category, in every episode. Or at least give a brief overview of my understanding of them. In this first episode we are going to start with ChatGPT, the insanely popular AI chat product right now, and Midjourney, a tool that creates images based off of text prompts.
I would guess that if you are listening to this, it is highly likely that you have already played with ChatGPT. I mean, of course you have, everyone is talking about it! All you have to do is go to chat.openai.com – and here is a major time saver that I can really appreciate – you don’t even need to “create an account” if you already have a Gmail account, which I assume many of you do – you just click “sign in with Google,” choose your Gmail account, and you are in. Easy.
So why is this text-based AI chat client all the rage currently? Well, basically we are talking about “AI for the masses” or “AI on the go” and honestly, this has been a pretty big disappointment up until now. Siri, Alexa and friends have been on the market for YEARS, and they can tell you the weather or pull up a recipe, or play a song, but most of the time I use these products, I am disappointed. You would think that Siri could switch my AirPods from “noise cancellation” to “transparency” mode, but you’d be wrong. And that’s using all Apple software and hardware! Try asking a follow-up question to your previous inquiry; it’s often misunderstood. It’s not a conversation. Well, boom, now that’s changed. Overnight.
You may have heard that you can ask ChatGPT to write you an essay, or summarize an article, or even re-write “Waiting for Godot” in the style of Quentin Tarantino (yes, I did that), but this is only the very beginning. This is like being handed the keys to a Ferrari and saying, “Wow, the door makes such a cool sound when it opens and closes,” and then walking away. You didn’t get a very deep experience with the product. Full disclosure, I have no idea what a Ferrari door sounds like, but if you start to spend some time with me, you’ll get used to my random strangely attempted metaphors.
Here are some other examples of what you can ask ChatGPT:
Brainstorming: this is huge. I mean one of the core benefits of all of this technology, in my opinion, is that we don’t have to stare at a blank screen – or tabula rasa if I am recalling my Latin phrases correctly – anymore. Now, you can always get thought starters. Like:
What are the top 5 themes and trends that I should explore as I start to write a column on [insert subject]?
What questions could I ask a panel at an event about [insert subject]?
Emails: there are specific AI tools focused on email writing, but you can start right with ChatGPT. Things like:
Write an email to an influencer asking them to promote [insert product or event]. Here are the product (or event) details..
Write an email to an existing customer asking them to take a survey and leave a Yelp recommendation. Have a humorous tone and offer a discount code for their next purchase.
Write a persuasive email to a customer who abandoned their shopping cart. Offer a discount and explain the benefits of completing their purchase.
Business: you can ask for lots of detail with questions like:
Develop a content marketing strategy for [insert business]. Include target audience, key themes, and suggested content for each stage of the customer journey.
Generate a Strengths/Weaknesses/Opportunities/Threats analysis on [X type of business]
What are three product ideas in [X sector] that could potentially be profitable within 4 years? Provide specific product details and explain how they will generate revenue.
Generate 4 detailed customer personas for [X type of business in Y location] – I asked this one for a pub style restaurant in Austin, Texas. The personas were well fleshed out and insightful.
Of course, as we are in the early stages here, some information may be inaccurate. As long as you don’t see the AI result as the final product, and more of an assistant that helps you along your journey to the final product, you will be blown away at what this technology can do.
I once asked ChatGPT if I gave it some plot themes and ideas for characters, would it be able to reply with a fully developed three act structure for a film? It said, “Let’s do this bro!” and I did. And it did. Was it amazing? No, but it wasn’t that bad either. And now that neglected Chrome tab that I’ve had open for over 4 years containing what I had hoped would become a screenplay eventually is suddenly invigorated with new life – I have a writer’s assistant if I want one. And this is not even mentioning all of the actual tailored-for-writers AI products that are already out there, which I hope to highlight in another episode.
Here’s the deal. For an incredibly long time, it has held true that you can have two out of three of the following: fast, cheap, and good. Just two. Fast and cheap, but not good, or cheap and good, but not fast, or fast and good, but not cheap. Well, this proverbial law of production is seriously challenged now. Fast? Check. It comes back with results almost immediately. Cheap? Check. It’s free! Who knows what future pricing of features and product iterations will be, but right now it’s free. And good? This is a matter of opinion, but let’s be real: the quality of responses is definitely not bad, that is undeniable. And it’s just going to get better.
I don’t want to ignore all of the incredible questions these technological leaps are presenting, from academia and education topics like test taking and plagiarism, to intellectual property, to its effects on labor and the future of work.. I mean the list goes on and on. Even saying, “it’s free!” leads to discussions of ultimate ownership of the end result, or how your input is being used in part to help train their models, and improve their product. You’re working too! But not for free; and it’s up to you to decide the value of what you get in return, and if you want to continue. My intent is to do an entire episode asking plenty of these questions. Please don’t expect many answers!
The truth is, generative AI is here, now, and it’s not going away. You can’t really put the toothpaste back in the tube, as they say. And the money sees where the puck is going and it’s moving there. Smart companies, big brands, are already embracing it. At the time of production of this episode, BuzzFeed recently announced that it will use AI to help create its content. What a no-brainer for quiz creation, listicles, and their other flavors of text. And on the venture capital side, a lot of money is leaving web3 and flowing into generative AI.
An old friend of mine, Josh Brown, recently wrote a blog post called, “ChatGPT is everything you wanted Bitcoin to be” – it’s a winner of a 3 minute read.
Like I said, starting with ChatGPT is simple. You go to chat.openai.com, sign in with Google, and start playing with it, merely by asking questions or asking it to do things, via text chat. And the examples I provided here are just scratching the surface.
Now let’s move on to Midjourney. While ChatGPT is a product of OpenAI, a company in San Francisco with several hundred employees which just got a $10 billion investment and major partnership with Microsoft – Midjourney is a tool from a company of the same name, also in San Francisco, with roughly a dozen employees. Just incredible.
Now, OpenAI has a text-to-image product as well. You have probably heard of it, it’s called DALL-E. And rounding out the big three of the image generators in early 2023 is something called Stable Diffusion, which I haven’t even used yet. It looks like it requires more local processing power and is less cloud-based, and I’m not set up for that.
Anyway, the thing that was obvious to me early on with these image generation tools is that there has emerged a whole new field of human interaction with computers – the prompt writing. When you start to look at some of the incredible images that are being generated, and then the prompts that yielded them, they aren’t just one sentence. They are often a full descriptive paragraph, and they have their own sort of jargon – like any professional field – with keywords and phrases that trigger certain visual styles. Prompt writing is a thing, and it will become an even bigger thing. There is a company called Lore Machine that as of this recording is hiring for several roles, and one of them literally has the title of “Chief Prompt Officer.”
Okay so let’s get into how you start to use Midjourney. Where ChatGPT was as simple as signing in with your Google account and then starting to type text, Midjourney is a little more involved. But it’s not that bad and I can walk you through it quickly.
First off, Midjourney uses Discord as its interface. If you’ve never used Discord, I’m going to quickly explain how this is done. I know a new tool can be a challenge, and I was there a few years ago as I was heavily exploring the world of NFTs and NFT artwork. The NFT social party is largely on Twitter and Discord, and every time I went into the unfamiliar world of Discord, I was sure I was going to click something that emptied my Metamask wallet or transferred my NFTs to a thief. It was terrifying. But this is not crypto and you don’t have to worry about that with Midjourney, everyone is just making imagery. Of course, be safe out there with any social media tools and platforms. Be sure to use common sense when interacting with other people.
In broad terms, Discord is also similar to Slack, if you’ve used that platform. At the end of the day it is a collection of themed areas that you can join (or be invited to join), each containing a bunch of chatrooms or channels. I look at Discord “servers” as islands or villages, and when inside each one I can jump into dozens of chatrooms related to the main subject. Discord started mainly as a social platform for gamers, but now there are around 7 million servers, and many are not related to gaming.
So, when you go to midjourney.com and click “join the beta” – it will lead you to a Discord account creation if you don’t already have one. If you do, it will invite you to join their server. On their server and website are plenty of resources on how to get started. But here is the quick version. You start by typing “/imagine” and then your text prompt, in a chatroom, and the Midjourney bot will respond in the chatroom with a square image that contains four low-resolution images of its response. You can then regenerate images based off of one of the four, or upscale one of them. There are lots of chatrooms for newcomers that start with “newbies” – followed by a number, and you can jump into any one of them and get started. It’s free to mess around for a while, but once you use a certain amount of computing time, you will need to sign up for one of their plans to continue. The basic plan is $10 per month.
Here’s a quick tip that I found really useful, since these chatrooms are usually pretty busy and there is a constant stream of prompts and imagery flowing. It can sometimes be difficult to follow where your responses and activity went, so you can just add the Midjourney bot to your own Discord server.
Yes, you have a Discord server. Everyone has a Discord server, and if you’re like me, it was just a private, boring area that I neglected. I never had any intentions of starting up my own area, inviting people, and managing a community. But, you can use this quiet space as your own personal Midjourney sandbox. When you are inside the Midjourney Discord server, click the tiny little “people” icon in the upper right, which pulls out a panel where you see a subset of all of the members – but also you will see the Midjourney bot. Click on it, and then there is a big purple button that says, “add to server” – click that and add it to your server – and now you have your own personal Midjourney play area. You just type “/imagine” into any of your chatrooms followed by a prompt and away you go. I still like to go back to the Midjourney server, where I can see what other people are doing, but I find having my own area much easier when generating images.
The main image I used for this podcast, a dalmatian sitting inside a light blue dome-shaped room, was generated with Midjourney. Here was my prompt:
"a room with a slight blue color filled with natural sunlight from a skylight. the room has no corners, it is in a dome. there is a dalmation dog sitting in the middle of the room."
That’s it. Not even a paragraph-long prompt with keywords. I’ve got some work to do before I can apply for that Chief Prompt Officer position, but the result was impressive, in my opinion. I only did one variation and then one upscale.
And what do I mean by that? Well, after you generate the low-resolution square image divided into four quadrants with the initial four options, there will be eight little buttons to the right: U1, U2, U3, and U4 – and V1, V2, V3, and V4. The 1, 2, 3, and 4 correspond to the individual images in the quadrants. The U buttons upscale, generating a larger version of the selected image and sprinkling more magic on the details. The V buttons create variations of the selected image. Creating a variation generates a new image grid with four images similar to the chosen image's overall style and composition.
There’s much more, but these are the basics to get you started. If you’re having writer’s block on what to ask Midjourney to imagine, try asking ChatGPT for help!
Well, if you’ve made it this far, thank you. Looks like I clocked in around 17 minutes; I hope you found it time well spent. Oh, and did I use ChatGPT to help me write this episode? Of course! But not that much, honestly. I’m doing my best to not sound like I’m reading, but this is a script that I wrote as best as I could in my own speaking voice. ChatGPT can’t do that for me. But, now that I’ve written one, I could try asking for another episode in this style on a new topic. There are even voice synthesizing AIs that can recreate my voice saying anything, based on a small audio sample. I should be able to crank these out pretty quickly if you think about it! I mean, what’s left to do but proofread to make sure deepfake me isn’t saying something I don’t like before publishing? It’s getting real crazy, real fast, people. In my next episode, I hope to embrace the crazy and start asking some difficult questions. Thanks again for listening, and buckle up!