Day 1
Welcome to the AI Jungle.
In keeping with the theme of learning out loud, here’s where my thinking about AI an digital marketing is at as of this specific day in 2023.
Like a lot of us, or at least those of us who will admit it, the quality of ChatGPT, its limitations notwithstanding, surprised me. Of course, I knew AI was coming for my job, I just didn’t think it was here, yet. Maybe in a few years, I rationalized. I’ll be retired. It won’t matter.
Well at least as of today, I still think it’s not here in terms of replacing me with all my experience, but what’s changed is my perception of when that event might happen and exactly what skills are going to be less in demand. But the window has closed from “sometime in the next decade” to “ near certainty in the next 3 years.” And that’s why I’m here.
Will any of us forget what it felt like the first time we put a prompt into ChatGPT and watched it write a grammatically correct, typo-free, properly structured blog post about our area of expertise in less than 60 seconds as we watched? I was horrified.
And a second later I was entranced.
The last time I had this feeling was when someone showed me the first iPhone. I dragged people into Apple stores to show them: the world is now different. I wasn’t sure how, but I knew everything would change. It won’t change in the same way. This isn’t a consumer play. This is about how AI gets folded into the tools we use every day, and how it powers tools we haven’t even imagined yet. I have to give credit to my friend Adam Nash for pointing me in the right direction. He explained to me that companies are, as we speak, adding AI to their tools every day; and my expertise in cutting through the bullshit hype to focus on ROI was going to be more in demand than ever.
Fortunately for me, I’ve got more time on my hands these days. I mean it’s not fortunate that I had breast cancer, but the space it’s given me to stop and reset some things that were not working for me personally and professionally has been invaluable. Hate the reason, love the opportunity. Enough of that, onwards…
My thoughts this week on AI (mostly focused on NLP at the moment) for Marketing:
ChatGPT gets an inordinate amount of attention vs the entire sector. In that way, its relationship to AI is much like that of Bitcoin to Blockchain. It’s important to not conflate ChatGPT with all of AI, or for that matter, machine learning. ChatGPT has competition from companies like Copy.ai, Jasper.ai, and more. And as we speak companies like Grammarly and I’m sure Microsoft are building AI into their word processing and editing software as well. One of the reasons I’m digging into this is that I want to be able to advise my clients on the exponential number of new choices that are about to be in front of them, which will make the problem of choosing marketing stacks that much harder.
To call these early tools AI is an overstatement. My friend Jim Sterne has a fantastic expression: “Machine learning is built in Python; Artificial Intelligence is built in PowerPoint.” That doesn't make these tools “vaporware,” but it does remind me to remind you that we aren’t anywhere close to replacing all the humans. These are rudimentary tools, which of course are about to get better, rapidly. Humans will still be required for a long time, yes even in writing and editing.
For now at least, never ask ChatGPT questions for which you do not already know the answer, or can easily verify. Because ChatGPT is an average of the internet, what it spits back is usually vague and generic. It’s not only wrong frequently; it’s wrong in insidious and subtle ways.
As a demonstration, here’s a query I recently ran: “Who is Melinda Byerley?” This is the entire answer. After you read that, I’m going to break it down for you.
Ok, it sure sounds authoritative, right? It’s even flattering, compared to the SEO results on my name, driven by hard right-wingers with their panties in a twist about some of my tweets.
But let’s drill in:
The first paragraph is pretty good. It’s not as compelling or descriptive as a human would write and there are some minor inaccuracies. For example, my company’s name is Fiddlehead, but we changed that after ChatGPT stopped ingesting the internet in 2021. Not a major flaw. Also, we don’t specialize in (early-stage) startups and small businesses, but this isn’t a deal killer.
Let’s move on to the second paragraph.
The paragraph starts out well enough. I have been mentioned in the publications above. The Inc magazine article isn’t accurate or flattering, and I’m amused at how ChatGPT sells it as an accomplishment. But it’s factually correct.
But everything after this statement is untrue. I have never spoken at SXSW or the Lean Startup Conference. I’ve never even attended SXSW! And I did not write the publication listed here.
The third paragraph is where it really gets weird.
For those who don’t already know, you need to know that I have for almost 15 years maintained what can courteously be called a spicy Twitter feed. I have strong opinions on politics, and I f
earlesslystupidlyfrequently share them out loud on my verified Twitter account. This has led to several dustups that right wing media has taken and run with, including dozens of obscure websites that tried to make me the character of the day on more than one occasion.
When I read the first sentence of this paragraph I laughed out loud. This was ChatGPT’s way of summarizing all the really horrible things that were said about me, especially the right’s use of the phrase “social justice warrior,” cleaned up and sanitized, and presented as innocuously as possible. Really I have to thank ChatGPT for showing me a much better way to describe myself than I ever could have imagined.
But it’s not accurate that I have been involved in several community projects. I have volunteer work I do, but it’s not something I write about publicly. To the best of my knowledge I haven’t donated to Planned Parenthood or the ACLU (I have related causes where I focus my giving.)
The current danger of ChatGPT is using it when you don’t know or can’t verify when the answer is wrong. Proceed with caution.The big opportunity with ChatGPT and other AI tools is what companies do with it on their own websites and products, trained on their own proprietary data. For example, big companies will train text tools on their own websites; enabling the tools to spit back copy that is in a brand’s tone and style. They are rolling these tools out already, most within weeks of ChatGPT’s release last November:
Microsoft has Azure OpenAI and will integrate with all its products.
Brands like Chase are using companies like Persado to augment their campaigns by generating better-performing taglines, more efficiently segmenting and targeting audiences, and more.
Domain experts are more important than ever. It’s time to dig in. The other night I sat down to dinner with a truly expert B2B copywriter and editor with over 40 years of experience writing truly amazing stuff. Understandably, writers are shocked and frightened about what they are seeing. I don’t see AI as an unmitigated benefit and I don’t see it as an unmitigated evil. It’s got the potential to be both, and those of us who know things should be comforted that our expertise is needed more than ever. Don’t you what to know what that brilliant writer thinks about the differences between Jasper.ai, copy.ai, and ChatGPt when it comes to writing marketing materials? I know I do. This goes for all our domains of knowledge. If someone’s going to work on a chatbot or AI/ML tool for marketing analytics that will interpret the data, they will need someone like me to help ensure that the tool isn’t giving the wrong advice.
I hope that I’m not coming off as some kind of AI cheerleader. I have concerns about the ethics of these tools and their impact, and I’ll be talking about them in future posts. At the same time, I can see that the future is here, and it’s not going to be stopped. If I can have an impact on how these tools are implemented and rolled out I want to have that impact.
Playing with Midjourney
Generative AI for images isn’t my field as much as natural language processing, but as so much of marketing is imagery, and because I do have some design background from my theater days, I plan to make regular exploration of visual design tools part of my content here. Here’s a further exploration of the prompt I used for the icon for this Substack, refined. I learned this week that artwork created via AI cannot be copyrighted, so as I said above already there’s a strong use case for visual artists to create iconography and content that can be copyrighted. For now, enjoy.
Prompt: extremely realistic fine details horizontal orientation curious happy ten-year-old girl with short dark hair and glasses in a bright forest with an owl
edit 3:30pm Friday, February 24, 2023: one of the founders of Netflix, Marc Randolph, posted Steve Jobs talking about the power of automation in a tweet that directly addresses the power of domain expertise layered on top of GPT3. Feels like things are coming together…
Let me know what you think about this content in the comments. Is it useful? Who do you follow to keep up with on this stuff? What tools are you using? I’m doing this mostly for my own learning, but if specific ideas and topics come up frequently, it can help me find angles on the ideas that I do have.
I was skeptical too, until I had that "holy shit" moment you described. It took me less than 2 months to fully embrace AI to write. This future is already here. Buckle up. Looking forward to where you go with this Substack.