Most of us think of Google when we think of algorithms and machine learning models.

And really, who can blame us? We are marketers and many of us are SEOs. we can’t help ourselves.

But there’s a lot going on inside and outside the Googleplex right now, making it all the more important that we keep up.

In this article, we’ll take a closer look at some new and exciting technologies. Where applicable, we’ll cover some of its current uses and then discuss where the technology is headed in the near future and how it will impact marketers.

So let’s dive in – let’s start with arguably my favorite ‘new release’.

1. Stable diffusion

Stable Diffusion is a text-to-image model built by Stability AI. Essentially, it lets you generate some nice images from your text prompts.

Because the model is open source and publicly available (on GitHub), you can easily get it and build a variety of tools and applications to suit your needs.

Here are some examples in action.

This shoe doesn’t exist.

And Johnny Depp has never done a photo shoot like this, nor has anyone taken the painstaking effort to create this in Photoshop. .

Prompt engineering is essentially playing around with different words, word order, and syntax to produce the type of image you want.

If you’re interested, you can try Stable Diffusion yourself here. You’ll need to create an account or authenticate yourself using Google or Discord, but it’s worth it. 😊

If you want to see Stable Diffusion in action with code (but you don’t have to write it), I made a Colab here.

Stable Diffusion has already been used to create images for advertising (I’ve used it myself) and websites (do you like the featured image in this article?), so the current use case is It consists of

There is already a PhotoShop plugin available for download here that allows you to easily integrate Stable Diffusion images directly into your creations.

obvious question

This raises some obvious questions, such as who owns the rights to the work. Images cannot be copyrighted as they are not actually yours and will soon be in the public domain.

What about the issue of creating an image of a person without their consent? What if they have a product? Worse, if I can’t own the copyright because it’s not mine, what liability do I have for other images that may be created?

I’m not going to go down the ethical rabbit hole with you here, but there are many things to consider.

Purely as a marketer, if you create an ad campaign with images generated by Stable Diffusion, your competitors might take the images and reuse them, and you can’t do anything about it.

down the road

Last spring saw the rise of the text-to-image model with the DALL-E Mini (now Craiyon). You can play around with that model here.

Stable diffusion is a leap forward. Assuming things continue in the same direction over the next few months and years, I predict a rapid evolution towards text-to-video generation, such as creating video tutorials from text.

Additionally, I expect to see automated WordPress plugins soon that create images for your site based on the surrounding content.

But perhaps more interesting are some of the commercial opportunities Sergey Galkin captures brilliantly in this video tweet.

It’s worth noting that OpenAI also created DALL-E 2. It’s definitely better quality, but it’s not open source, so it’s less versatile and more expensive.

2.GPT-3

The GPT-3 algorithm was developed by a team of researchers from various institutions. However, Geoffrey Hinton, Yoshua Bengio, Yann LeCun of the University of Toronto, and Andrew Ng of Stanford University have made significant contributions to the development of GPT-3.

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 would be able to create a model that better captures the meaning of the text.

GPT-3 is fine-tuned to improve performance on a specific task or set of tasks. For example, if GPT-3 is used for machine translation, it can be tweaked to improve its accuracy for that task.

Fun facts: An interesting fact about GPT-3 is that it was used to describe the “almost correct” three blocks of text above. This should give you an idea of ​​how it impacts your marketing. If you want to play around with it, you can do so here.

Additionally, you can use the GPT-3 system to power your ad copy generation.

Nearly two years ago, Search Engine Land covered the then-new PPC advertising and landing page copy creation tools. Well, those tools still exist, are being improved, and are still being used, one of which he recently secured $10 million in funding.

From a PPC perspective, it tends to work similarly to what we’ve seen before with Google Ads headline and description suggestions, but it can be more fine-tuned and the system is constantly improving.

obvious question

This leads to many questions about the future of content and content creation.

Google says it doesn’t like auto-generated content and that it’s considered spam because it violates its guidelines, but it’s devoted vast resources to technology designed to essentially do the same thing. (details below).

After all, Google makes guidelines, not laws, so the big question we have to ask is whether what we’re making provides the best (or at least better) user experience. That’s it. Unfortunately, even fine-tuned GPT-3 models are far from perfect at this point, and the content they generate needs fact-checking and frequent editing.

At the end of the day, it’s often as hard work as writing content yourself, but using GPT-3 can help you come up with ideas and information you might not have thought of on your own.

down the road

Will AI Take Over Writing? Not for the foreseeable future.

The advantage of us humans is that we can write about things we have never encountered before. We can create unique ideas based on observation and imagination. Machines can’t do that, so systems like GPT-3 have to come to terms with the content they create and the facts.

That said, some writes will soon be automated. Most copy of Google Ads he thinks will be automated within 5 years (like it or not).

I can’t see Google Ads announcing that you just provide a URL and budget, and from there it will generate ads and bid strategies, showing you about 20% of the data you need about what’s going on. please teach me. Inside the black box.

20% may be too much, but you can see what I’m aiming for.

However, we also need to take advantage and put more energy into our landing pages and experiences. We get help from a language model that helps us communicate with our customers (chatbot powered by GPT-3 or Meta AI’s public BlenderBot 3?) to help us research and create first draft content.

3. Mom

This was the model I had in mind when I mentioned earlier that Google was developing a system to create AI-generated content. MUM, along with other similar models to be developed in the coming months/years, will change dramatically in marketing methods alone, but where.

Let’s quote from Google’s MUM article.

“…MUM doesn’t just understand language, it generates it. By training 75 different languages ​​and many different tasks at once, we are able to capture information and knowledge of the world more comprehensively than previous models. MUM is multimodal, meaning it understands information across texts and images, and can be extended to more modalities such as video and audio in the future.

When you ask about hiking Mount Fuji, MUM understands you’re comparing two mountains, so elevation and trail information may be relevant. It’s also understandable that in the context of hiking, “preparation” could include things like fitness training and finding the right gear.

MUM is able to surface insights based on his in-depth knowledge of the world, so he noted that although both mountains are about the same elevation, autumn is the rainy season for Mt. Fuji, so waterproof jackets may be needed. I can emphasize. With pointers to helpful articles, videos, and images from across the web, MUM can also surface subtopics for deeper exploration, such as top-rated gear or best training exercises. ”

The key here is that MUM allows Google to collect information from different languages ​​and modalities and use this information to generate unique content/answers.

Yes, they display it in a good format with examples and suggest using it only to endorse articles and products, but in reality, they use it to craft answers.

After all, one of its capabilities is the ability to understand information across languages. Getting article recommendations in a language I don’t speak doesn’t help me much.

So basically they use the information they are collecting and present it as a complete answer to the end user. Collect from enough sources, but no one to cite.

down the road

The big “future” with this is understanding that as it is fully deployed in the environment, there is less room for organic consequences. Created by Google based on their knowledge of the whole.

No attribution. No clicks.

Organic isn’t going away, SEO isn’t going away (sorry naysayers). However, the structure changes dramatically.

Imagine a world where search results consist of answers that contain only secondary and tertiary links to resources. Think of a LamDA/chat world where each outcome means an engagement rather than the end of a story. Engagement intended to fulfill a user’s intent rather than simply answering a question.

Imagine the marketing opportunities that will arise. Weave content into new locations. Trigger conversions by displaying ads at the right time during discussions.

Don’t get me wrong, it’s not just sun and roses. My heart goes out to publishers and people who are getting less exposure and whose content is their primary product.

what’s next?

When it comes to marketing and its future, there are many other machine learning models to explore. Some might even say the best is yet to come.

In the next article, we will discuss Augmented Reality and the Metaverse. What directions they might take, what they need to do to prepare for this brave new world (or is it non-world?), and the machine learning engineers working to build this reality. I will discuss some points from the interview.

The opinions expressed in this article are those of the guest author and not necessarily Search Engine Land. Staff authors are listed here.

About the author

Dave Davies is the Lead SEO at Weights & Biases, a machine learning operations company. He started his SEO in the early 2000s and co-founded his Beanstalk Internet Marketing in 2004 with his wife Mary. He hosts weekly podcasts, speaks regularly at industry-leading conferences, and is proud to be a regular contributor here at Search Engine Land.