Almost Timely News: Getting Started With Generative AI 101 (2023-07-02)
Almost Timely News: Getting Started With Generative AI 101 (2023-07-02) ::?View in Browser
Content Authenticity Statement
95% of this newsletter was generated by me, the human. AI generated content appears in the first section in the form of a prompt’s response and an AI generated image.
Watch This Newsletter On YouTube ??
What’s On My Mind: Getting Started With Generative AI 101
A friend was telling me yesterday that her therapist - her THERAPIST - was suggesting she “get good at this AI stuff”, in the context of a discussion about career. Imagine that. Naturally, my friend - as well as many, many other folks - have said, “Okay, so where do I start?”
Where do you start? There are a ton of different frameworks you can use to plot a journey through AI, but the one that makes the most sense for the average person is the why/what/how. For the average business, it’s the?Trust Insights 5P framework. Since this is in the context of one friend at a personal level, let’s use the personal one, and we can tackle the business one another time or in the?Trust Insights newsletter, INBOX INSIGHTS.
So, why/what/how. Why do you care about AI? Why SHOULD you care about it? What is AI? And how do you get started? Let’s dig into each of these three topics. We’re going to specifically address generative AI, which is the most accessible and useful form of AI for the average, non-technical person. Recall that there are three big categories of AI - prediction, classification, and generation; generation is what we’re talking about today.
Why should you care about generative AI?
Not because it’s the cool shiny object right now, or because your therapist told you to. Not because it helps businesses make stuff better, faster, and cheaper. Not even because it’s going to cost thousands, if not millions of jobs in the big picture. The primary reason to care about AI is a simple truth, across professions and industries. AI isn’t going to take your job. A person skilled with AI will take the job - or jobs - of people not skilled with AI.
Why specifically should you care? In general, generative AI is about making stuff, either net new stuff or derivatives of existing stuff. If any part of your work involves making stuff - from writing emails to putting together ads to composing songs - then getting a handle on what generative AI can and cannot do is critically important. You need to know what parts of your job you’ll still need to do (like showing up to meetings) and which parts AI can and should do (like writing up meeting notes from all those meetings).
Here’s a simple guideline: if a task is repetitive and involves creating something (like a weekly recap email to your boss), it’s a good candidate for AI to assist or outright do. Think about all the tasks you do at work. How many of them fit in this category? This is the first and most important thing to do. If you literally have nothing on your task list that fits in this category, then there might not be as much urgency to adopt AI, but it will be something you have to contend with eventually.
For example, Microsoft is rolling out its Copilot generative AI integration into Microsoft Office later this year. This brings up a plain language prompt in Office that allows you to do things like say, “Convert this spreadsheet into a written narrative” or “Make a slide presentation from this memo”, as well as more conventional generative tasks like “Help me write this email to the staff telling them they’re all fired”.
Even relatively straightforward tasks like writing an agenda for a meeting are fair game for AI to help you. Google’s Duet is the Copilot equivalent for Google Docs and Gmail. And AI will be in nearly every software package you use for every job. It’s already in tools like Adobe Photoshop, Hubspot’s CRM, Salesforce, Unity’s video game development engine, and so many more.
What exactly is generative AI?
Okay, so we understand the importance of generative AI. Now let’s talk about what the hell this stuff is. Generative AI comes in two flavors because of their fundamental architectures, transformers and diffusers. Transformers are found and used mostly in language generation, with software called large language models. When you use services like Google Bard or ChatGPT, you are using transformers. Diffusers are found and used mostly in image generation, with software called diffusion models. When you use services like DALL-E, Stable Diffusion, or Midjourney, you are using diffusers.
How these two architectures work is fairly complex, but here’s a simplified explanation. Let’s say we want to be able to make pizza. If we’re using transformers and large language models, the companies that make these models go out and eat a whole bunch of pizza. They try pizza from all over the world and in every variation they can find. They take notes on each pizza as they eat them. When they’re done, and done being very sick from overeating, they assemble their notes into a cookbook. That cookbook is the transformer - and when someone asks for a pizza, they can reference their notes and make a pizza that fits what someone asks for. This includes pizzas they’ve never heard of before, because they’re smart enough to understand if someone wants a gluten-free mushroom and popcorn pizza, they can still assemble it based on the logic of past pizzas they’ve tried. That’s how transformers work - they ingest a huge amount of text and then try to guess what words they should spit out based on the instructions we give and the text they’ve seen in the past.
If we’re using the diffusers model, the companies that make these models still go out and eat a bunch of pizza, but when someone asks for a new pizza, what they do is throw pretty much every ingredient on the dough and then refine it. They add stuff, remove stuff, change ingredients, change amounts, until they arrive at a pizza that most closely resembles the pizzas they’ve tried in the past. That’s why diffusers work really well with images; they start by throwing all the pixels into the mix and slowly refine it, adding and removing pixels until the image looks like what we asked for, like a dinosaur sipping on a cocktail and reading a newspaper.
Both models perform the same fundamental two tasks: comparison and generation, or more simply put, editing and writing/creating.
For example, diffusers in images can create net new images based on a prompt, like the dinosaur sipping on a cocktail and reading a newspaper. But they can also do tasks like inpainting, where they change part of an existing image, or outpainting, where they extrapolate the rest of an image from a portion you give them.
Transformers can generate new text like memos, blog posts, etc. as well as answer questions like, “Where in Prague can I get a really good steak?” with a high degree of success. They can also perform tasks like summarizing large amounts of text, rewrite text, extract information from text, and classify text by attributes like sentiment or tone of voice.
Generally speaking, AI models are better at tasks that are editing tasks like inpainting or summarizing text because there’s less data needed to generate the results than there is with creative tasks like writing a new blog post or making a brand new image from a prompt. As you evaluate your list of tasks that you’d want to use AI for, think about whether the task is an editing task or a creating task. Writing an email newsletter each week is a creative task (though I still write this one by hand, because I haven’t had time to fine tune a model on my exact voice). Summarizing the meeting notes from a client call is an editing task.
So now you’ve got sort of a basic decision tree. Are you working with text or images? And are you doing editing or creating? That leads us to the third question: where do we get started?
How do you get started with generative AI?
Inevitably, the first question people ask once they wrap their heads around AI is which tools they should be using. Imagine, once you learn the existence of and utility of cooking, immediately starting by asking which appliances you should be using. To some degree, that makes sense, but it makes more sense to learn the broad types of cooking and then understand the ingredients, tools, and recipes for those types. Running out to buy a blender with no idea of what you’re going to make is going to yield unpleasant results if you then realize all you have in the refrigerator is chicken wings.
By spending time cataloging the tasks you do as image or text-based, and then whether you are doing editing or creating tasks, you are setting the groundwork for being successful with AI. There are hundreds of new AI vendors popping up every week, and for the most part, they all do more or less the same things. Everyone’s got the same foundational models to start from that they’ve done some tuning on, or they’re just using someone else’s model. Some services have a better UI than others, some have better customer support than others, but they are all using some form of transformers or diffusers if they’re offering generative AI.
That means that at least early on in your AI journey, you can ignore the vendors and the hype while you get your feet wet. You’re not missing out on anything critical while you master the basics. And where do you master the basics? You start with the free foundational tools.
For transformers and large language models, the best place to start as long as you’re not working with sensitive or confidential information is?OpenAI’s ChatGPT.
For image generation, the best place to start is?Microsoft Bing’s Image Creator.
These two tools have the lowest barrier to entry, the lowest cost, and have some of the best basic capabilities.
Once you’re successful with these tools, then start looking at more specialized tools, vendors, and platforms.
The first skill you’ll need to learn is prompt engineering, which is essentially just programming these software models using plain English language.
For transformers and large language models, the general template you want to use is role / task / background / action.?Download my cheat sheet here for more details on why. For example, if I wanted ChatGPT to write a memo telling staff not to microwave fish in the breakroom microwave, I might prompt it like this.
You are an executive assistant. You know how to communicate diplomatically, handle difficult situations, manage confrontation, set expectations. Your first task is to write a memo asking staff not to microwave fish in the breakroom microwave. Some background information: fish is very difficult to clean the smell. Fish dishes can be heated using the induction plate in the breakroom. Many staff do not enjoy the smell of fish, and it can cling to other foods. Be considerate of your fellow workers. Write the memo in a professional tone of voice.
You put this into ChatGPT, inspect the results, and either tweak the prompt or just polish the results by hand:
For diffusers and image generation, prompts look a lot more stilted because of the way diffusers work. They almost read similar to how captions read on famous artworks, like this one:
Title: The Abduction of Europa Creator: Rembrandt Harmensz. van Rijn Date Created: 1632 Physical Dimensions: w78.7 x h64.6 cm Type: Painting Medium: Oil on single oak panel
If you were to write a prompt for a system like Bing Image Creator, you might write something like:
A redheaded woman riding across a river on a white horse while local villagers look on in shock from the riverbank, oil painting, Renaissance, in the style of Rembrandt, highly detailed, finely details, oil on oak panel
Here’s what the Bing Image Creator made:
In general, for image generation, you write the subject first with as much detail as you can manage, following by the format, then the style with as many relevant modifiers (like oil on oak panel or 35mm film) after. Why such a weird format? Diffusers were basically trained on captions of images, including those of artworks. Thus, it’s no surprise that prompts formatted similar to how artworks are described tend to work well.
Your next step is to take your task list of highly repetitive tasks and start writing prompts to see how to accomplish those tasks with generative AI.
Obviously, there’s quite a bit more we could cover and absolutely absurd amounts of detail we could go into about all the technologies, use cases, dangers, and implications,?many of which are in my talk about generative AI, but this is a good starting point, a good way to get going.
Commercial plug: If you’re really interested in talking shop about AI,?come hang out with me in Cleveland at the Marketing AI Conference, MAICON, July 26-27. Use discount code TRUST150 to save $150 on the ticket price.
Got a Question? Hit Reply
I do actually read the replies.
Share With a Friend or Colleague
If you enjoy this newsletter and want to share it with a friend/colleague, please do. Send this URL to your friend/colleague:
ICYMI: In Case You Missed it
Besides the newly-refreshed?Google Analytics 4 course I’m relentlessly promoting (sorry not sorry), I recommend the piece on why you need a prompt library for AI.
领英推荐
Skill Up With Classes
These are just a few of the classes I have available over at the Trust Insights website that you can take.
Premium
Free
Get Back to Work
Folks who post jobs in the free?Analytics for Marketers Slack community?may have those jobs shared here, too. If you’re looking for work, check out these recent open positions, and check out the Slack group for the comprehensive list.
Advertisement: Bring My AI Talk To Your Company
I’ve been lecturing a lot on large language models and generative AI (think ChatGPT) lately, and inevitably, there’s far more material than time permits at a regular conference keynote. There’s a lot more value to be unlocked - and that value can be unlocked by bringing me in to speak at your company. In a customized version of my AI keynote talk, delivered either in-person or virtually, we’ll cover all the high points of the talk, but specific to your industry, and critically, offer a ton of time to answer your specific questions that you might not feel comfortable asking in a public forum.
Here’s what one participant said after a working session at one of the world’s biggest consulting firms:
“No kidding, this was the best hour of learning or knowledge-sharing I’ve had in my years at the Firm. Chris’ expertise and context-setting was super-thought provoking and perfectly delivered. I was side-slacking teammates throughout the session to share insights and ideas. Very energizing and highly practical! Thanks so much for putting it together!”
Pricing begins at US$7,500 and will vary significantly based on whether it’s in person or not, and how much time you need to get the most value from the experience.
What I’m Reading: Your Stuff
Let’s look at the most interesting content from around the web on topics you care about, some of which you might have even written.
Social Media Marketing
Media and Content
SEO, Google, and Paid Media
Advertisement: Google Analytics 4
Believe it or not, July 1st, 2023 - and Google’s shutdown of Universal Analytics in favor of Google Analytics 4 - is in less than THIRTEEN calendar days. This means that in THIRTEEN days, you will no longer be able to capture data in Universal Analytics - it will just stop collecting data. If you haven’t already switched over, it’s urgent you do so right now. So, let’s get you moving.
Tools, Machine Learning, and AI
Analytics, Stats, and Data Science
All Things IBM
Dealer’s Choice : Random Stuff
Advertisement: Ukraine ???? Humanitarian Fund
The war to free Ukraine continues. If you’d like to support humanitarian efforts in Ukraine, the Ukrainian government has set up a special portal, United24, to help make contributing easy. The effort to free Ukraine from Russia’s illegal invasion needs our ongoing support.
How to Stay in Touch
Let’s make sure we’re connected in the places it suits you best. Here’s where you can find different content:
Events I’ll Be At
Here’s where I’m speaking and attending. Say hi if you’re at an event also:
Events marked with a physical location may become virtual if conditions and safety warrant it.
If you’re an event organizer, let me help your event shine.?Visit my speaking page for more details.
Can’t be at an event? Stop by my private Slack group instead,?Analytics for Marketers.
Required Disclosures
Events with links have purchased sponsorships in this newsletter and as a result, I receive direct financial compensation for promoting them.
Advertisements in this newsletter have paid to be promoted, and as a result, I receive direct financial compensation for promoting them.
My company, Trust Insights, maintains business partnerships with companies including, but not limited to, IBM, Cisco Systems, Amazon, Talkwalker, MarketingProfs, MarketMuse, Agorapulse, Hubspot, Informa, Demandbase, The Marketing AI Institute, and others.?While links shared from partners are not explicit endorsements, nor do they directly financially benefit Trust Insights, a commercial relationship exists for which Trust Insights may receive indirect financial benefit, and thus I may receive indirect financial benefit from them as well.
Thank You
Thanks for subscribing and reading this far. I appreciate it. As always, thank you for your support, your attention, and your kindness.
See you next week,
Christopher S. Penn