When many of us think of algorithms and machine learning models, we think of Google.
And really, who can blame us? We’re marketers, and many of us SEOs. We can’t help ourselves.
But there is a lot going on in and out of the Googleplex right now, and it’s becoming increasingly important that we keep up.
In this article, we’ll dive into some new and exciting technologies. We’ll cover some of the current uses where applicable, then move on to discuss where I see the technology going in the near-to-mid future and how it’ll impact marketers.
So let’s dive in – starting with arguably my favorite “new release.”
Stable Diffusion is a text-to-image model built by Stability AI. In essence, with it, you can generate some amazing images from text prompts.
The model is open source and public, meaning you can get your hands on it easily (on GitHub) and build a variety of tools or applications to suit your needs.
Here are a couple of examples of it in action.
And Johnny Depp never did a photo shoot like this, nor did anyone put in the painstaking effort to create this in PhotoShop. In fact, it only took me a few minutes of prompt engineering to create.
Prompt engineering is basically just playing around with different words, word orders and syntax to generate the type of image you want.
For those interested, you can play with Stable Diffusion yourself here. I should note, you do need to authenticate yourself, either by creating an account or with Google or Discord, but it’s well worth it. ????
If you’d like to see Stable Diffusion running with the code (but without having to write the code), I’ve created a Colab here.
Stable Diffusion is already being used to create images for ads (I know… I’ve used it myself) and websites (did you like the featured image for this article?) so the current use case makes itself.
There’s already a PhotoShop plugin you can download from here to integrate Stable Diffusion images directly into your work more easily.
This brings up some obvious questions such as who owns the rights to the work? It turns out, you can’t copyright your images because they’re not really yours and immediately become public domain.
How about the issue with me having created an image of a person without their consent? What if I had them holding a product? Or worse, if I can’t own the copyright as it’s not mine, how responsible am I for other images that might be produced?
I’m not going to head down the ethical rabbit hole with you here, but there is a lot to consider.
Thinking purely as a marketer, if you build your ad campaign on Stable Diffusion-generated images, they can be taken and reused by your competitors and there’s nothing you can do about it.
Last spring, we saw the rise of text-to-image models with DALL-E Mini (now Craiyon). You can play around with that model here.
Stable Diffusion is a leap forward. Assuming things continue to progress along the same line in the months and years ahead, I predict that we’ll evolve quickly into video generation from text, including the creation of video tutorials from text instructions.
Additionally, I imagine we’ll soon see automated WordPress plugins that create images for the site based on the surrounding content.
But more interesting perhaps are some commercial opportunities Sergey Galkin captures brilliantly in this video tweet:
It’s worth noting that OpenAI has also produced DALL-E 2 which is arguably superior in quality, but it’s not open source and thus less versatile and more expensive.
The GPT-3 algorithm was developed by a team of researchers from a variety of institutions. However, some of the key contributors to the development of GPT-3 include Geoffrey Hinton, Yoshua Bengio, and Yann LeCun from the University of Toronto, as well as Andrew Ng from Stanford University.
GPT-3 was designed to improve the performance of natural language processing models. The developers hoped that by using a larger and more diverse training set, they could create a model that would better capture the meaning of text.
GPT-3 is fine-tuned to improve its performance on a specific task or set of tasks. For example, if GPT-3 is being used for machine translation, it may be fine-tuned to improve its accuracy on that task.
Fun fact: An interesting fact about GPT-3 is that it was used to write the three blocks of text above which are "mostly right." That should give you an idea of the impact it will have on marketing. If you want to play around with it, you can do so here.
Additionally, GPT-3 systems can be used to enhance ad copy generation.
Almost two years ago, Search Engine Land covered what was then new PPC ad and landing page copy creation tools. Well, those tools are still around, improved, and still in use. One of them recently secured $10 million in funding.
From a PPC standpoint, they tend to work similarly to what you’ll have seen before in the suggested headlines and descriptions in Google Ads, but you can fine-tune them better, and the systems are constantly improving upon themselves.
This leads to a number of questions about the future of content and content creation.
Google has said they don’t like automatically generated content, and that it’s considered spam as it violates their guidelines, however, they themselves put massive resources into technologies that are fundamentally designed to do the same (more on that later).
At the end of the day, Google creates guidelines and not laws so the big question we have to ask is whether what we are producing provides the best (at least better) user experience. Unfortunately, at this time even fine-tuned GPT-3 models are far from perfect and the content they produce needs to be fact-checked and often edited.
At the end of the day, it can often be as much work as just writing the content yourself – though using GPT-3 can prove useful in surfacing ideas and information that you may not think of yourself.
Will AI take over writing? Not for the foreseeable future.
The advantage we humans have is that we can write about that which we haven’t encountered before. We can create unique ideas based on our observations and imaginations. Machines can’t do that, so systems like GPT-3 need to encounter content and facts to create from.
That said, some writing will be automated soon. I suspect most Google Ad copy will be automated within five years (like it or not).
Tell me you can’t see Google Ads announcing you now just give them a URL and a budget, and they’ll take it from there, generating ads and bid strategies and showing you about 20% of the data you want on what’s going on inside the black box.
Maybe 20% is too generous, but you get where I’m going.
This all said, we’ll be getting advantages at the same time, and will be left to put more energy into our landing pages and experiences. Getting assistance from language models that help us communicate with our customers (GPT-3 powered chatbots or maybe Meta AI’s publicly available BlenderBot 3?) and help with research and first draft content creation.
When I mentioned above that Google is developing systems to create AI-generated content, this was the model I was thinking of. MUM, along with other similar models that will be developed in the coming months/years will dramatically change now just how we market, but where.
Let’s take a quote right from Google’s write-up of MUM:
"… MUM not only understands language, but also generates it. It’s trained across 75 different languages and many different tasks at once, allowing it to develop a more comprehensive understanding of information and world knowledge than previous models. And MUM is multimodal, so it understands information across text and images and, in the future, can expand to more modalities like video and audio. Take the question about hiking Mt. Fuji: MUM could understand you’re comparing two mountains, so elevation and trail information may be relevant. It could also understand that, in the context of hiking, to “prepare” could include things like fitness training as well as finding the right gear. Since MUM can surface insights based on its deep knowledge of the world, it could highlight that while both mountains are roughly the same elevation, fall is the rainy season on Mt. Fuji so you might need a waterproof jacket. MUM could also surface helpful subtopics for deeper exploration — like the top-rated gear or best training exercises — with pointers to helpful articles, videos and images from across the web."
The big takeaway here is that with MUM, Google can collect information from various languages and modalities, and use this information to generate its own content/answer.
Yes, they’re displaying it in their examples in a nice format and suggest they’ll just use it to recommend articles and products – but in truth, they will use it to create the answers.
After all, one of the features is being able to understand information across languages. It’s hardly useful to me to have an article recommended in a language I don’t speak.
So, fundamentally they will be using the information they’re collecting and presenting it to the end user as a complete answer. Collect from enough sources, and there’s no one to cite.
The big “down the road” on this one is understanding that as it deploys into the environment in full force, there will simply be less room for organic results. Featured snippets will no longer be sources from a single authority but rather created by Google, based on their knowledge of the world at large.
Organic won’t go away and SEO is not dead (sorry to the naysayers) – but the structure will change dramatically.
Picture a world where the search result is constructed of answers with only secondary and tertiary links to resources. Think of a LamDA/chat world where each result is meant to be an engagement rather than the end of the story. An engagement meant to draw the user to fulfilling their intent, rather than just answering a question.
Imagine the marketing opportunities that will come from this. Weaving your content into new locations. Having your ads show up at just the right time in the discussion to trigger conversions.
Don’t get me wrong, it’s not all sunshine and roses. There will be less exposure, and I genuinely feel for publishers and people for whom content is the primary product. But for those selling products and services and who can adapt quickly, there will be a lot of opportunities.
When it comes to marketing and its future, there are a lot more machine learning models to explore. Some might even say the best is yet to come.
In my next piece, I’ll be exploring augmented reality and the metaverse. We’ll discuss some likely directions for them to take, what you need to do to prepare for this brave new world (or is it unworld?), and some takeaways from interviews with the machine learning engineers working to build this reality.
Opinions expressed in this article are those of the guest author and not necessarily Search Engine Land. Staff authors are listed here.