Back to Top

Tag Archives: AI

Microsoft Previews New ‘Designer’ Tool Which Enables AI Art Generation for Marketing Campaigns

Updated on by

AI-generated art is set to cause a major shake-up in digital marketing, with businesses able to use AI-generated images free of charge, which will eventually change the game for stock image providers, artists and more.

And Microsoft’s not waiting around for the revolution to happen. Today, Microsoft has shared a preview of its coming Microsoft Designer platform, which will provide a simple way to create content for your promotions based solely on text prompts.

What’s more, as you can see in this preview, the app will also incorporate DALL-E, the AI-based image generator that can take your basic text descriptions and turn them into art, within seconds, for use in your projects.

As per Microsoft:

With Designer, there’s no need to spend time building cards or social media posts from scratch, and you no longer need to search through thousands of pre-made templates. Designer invites you to start with an idea and let the AI do the heavy lifting. For example, with ’start from scratch’ within Designer, you can simply describe an image you want to see, and the app does the work for you to create something totally unique.”

Which seems good, and helpful in many respects. But also feels wrong?

As we outlined recently, at present, copyright laws don’t cover AI-generated art, because technically, the images being created through these apps have never existed before a user has generated them. These apps do source visuals from across the web, and many of those would likely be from commercial artists and platforms. But because it’s only sampling each image, then re-forming those elements into something new, that’s not a violation of copyright – though there are some questions around images of real people and/or public figures.

But in general, these visuals are fine for commercial use. Some people are even selling their original AI creations for profit, even though many do look a lot like other artworks that you would have to pay for online.

They can also look a little weird, a little distorted – but run enough text prompts through the system are you’re bound to eventually come upon an image that suits your needs.

Given this, it makes sense for Microsoft to jump on board the AI art shift, and you can expect other tech players to follow suit. But it does feel like a conflict, in potentially screwing over artists who are going to lose income as a result.

Take, for example, the use of named prompts in AI art generation, like this from Byteside, which is an image of ‘money tree in the style of Monet’ (created in Midjourney).

AI image example

That looks pretty good, right? And it does have a distinct Monet feel to it. And Monet, of course, is not around to be annoyed by this – but what about when such prompts are used to create art in the style of actual living artists, who are going to lose out as a result?

Fantasy artist Greg Rutkowski, for example, says that his art is often being sourced by these AI tools to create similar style works.

AI art example

As you can see in this example, the details are not quite right, but by using a prompt like ’wizard with sword and a glowing orb of magic fire fights a fierce dragon Greg Rutkowski’ in one of these AI generators, you can come up with similar looking visuals, based directly on Rutkowski’s style.

If you are interested in original article by Andrew Hutchinsin you can find it here

Posted in Blog | Tagged , , | Leave a reply
working-with-ai-62bc3ea50a305-sej-760x400

4 Ways To Use AI Right Now In Your Marketing Program

Updated on by

Learn how to automate all kinds of tasks with AI and use it as an assistant to level up your marketing activities.

In my last article, we saw that artificial intelligence is getting better and better at answering our questions, regardless of the subject matter or the sector.

Using GPT-3 technology, I demonstrated that an AI can successfully pass an SEO multiple-choice test as well as solve SEO case studies.

There are other, even more advanced, technologies, such as Deepmind’s Gopher, which outperform GPT-3 in the following fields: Humanities, social sciences, medicine, science, and math.

The following graph highlights the accuracy of the answers provided by Gopher, UnifiedQA, GPT-3, and a human expert.

Depending on the subject, we can see the narrow gap between the level of the AI and that of an expert.

This also suggests that the AI could potentially exceed the level of a non-expert person.

Performance on the Massive Multitask Language Understanding (MMLU) benchmark broken down by categoryImage from Deepmind, June 2022

Presently, AI can be a useful supporting resource for many marketing topics.

Let’s take a look at how to work with AI and more importantly, how to integrate it into your business.

4 Ways To Interact With An AI Tool

We will focus on the main methods of interaction to successfully set up the best AI-human combination.

1. Using Your Web Browser

The most interesting kind of interaction is to plug the AI into your back-office or into your web browser with a Chrome plugin.

There are numerous potential applications, as you will be able to help your users with advanced projects like document classification, writing assistance, meta-tag generation, text extraction, and even suggesting new topics.

You can connect the AI to your tools via a simple Javascript call asking it to perform specific tasks.

Below is an example of a JS integration with GPT-3.

The example is oversimplified to show you that with 20 lines of code and the appropriate instructions, you can easily connect a language model like GPT-3.


var xhr = new XMLHttpRequest();
xhr.open("POST", 'https://api.openai.com/v1/engines/text-davinci-002/completions');

xhr.setRequestHeader("Content-Type", "application/json");
xhr.setRequestHeader("Authorization", "Bearer sk-RkXXXXXXXXXXXXXXXXXXXXXXXX");

xhr.onreadystatechange = function () {
if (xhr.readyState === 4) {

if (xhr.status==200){
var data=xhr.responseText;
var jsonResponse = JSON.parse(data);
answerGPT3 = jsonResponse["choices"][0]['text'];
console.log(answerGPT3);
}
else {
console.log('API error');
console.log(xhr.responseText);
}
}};


var prompt = "List 50 concepts about … "

var data = `{
"prompt": "PROMPT",
"temperature": 0,
"max_tokens": 256,
"top_p": 1,
"frequency_penalty": 0,
"presence_penalty": 0
}`;

data = data.replace('PROMPT', prompt)

xhr.send(data);

If your back-office supports Javascript and you have seasoned developers, integrating advanced AI features has never been easier.

In order to test this initial connection, create a Chrome plugin to evaluate an AI on its capacity to correctly respond to certifications designed for experts.

To do this, use a Chrome plugin that allows optical character recognition to capture any kind of text.

Then, leverage a separate Chrome plugin that modifies a page’s CSS  to make the page as understandable as possible.

In an example exercise, we assigned each answer of a multiple-choice test to a corresponding letter.

Then, with the previous 20 lines of code, we sent the instructions to the AI to generate the results in a text field.

With this program, called “Asimov’s tests,” the AI managed to pass several certifications.

I then tested the AI on the subject of medical science and it attained scores of more than 60%, without any previous training in a specific discipline.

This confirms that by choosing your subject correctly, the results produced by the AI can strongly help your teams to improve day-to-day work.

SEO assessment from LinkedInImage from LinkedIn, June 2022

2. Using Your Data Visualization Tools

Over the past few months, tools that generate documentation or facilitate code writing have started to appear.

One remarkable use case is to simply generate dashboards or SEO tools with instructions.

There are now open source tools like Streamlit that have very advanced components in Data Visualization or Data Manipulation.

By providing the appropriate instructions, it is easy to request the generation of an app that interacts directly with your data.

For example, you can generate a web application with a complete interface and functional code.

This practice is quite recent, because we use language models that are exclusively fed with computer code. And again, the results are quite impressive.

In the following graph, you can see all the most popular code generators and the data with which they were created.

  • CodeParrot: 50 GB.
  • GPT-3 Codex: 159 GB.
  • InCoder: 216 GB.
  • PolyCoder: 249 GB.
  • AlphaCode: 715.1 GB.
  • CodeGen: 1.38 TB.

It is possible to generate applications in many languages; the main ones are Java, C, JS, and PHP.

most popular code generatorsImage via Huggingface.co, June 2022

Whether you’re a beginner or an expert, I encourage you to give it a try – as the AI can create your SQL query or your graphs in a matter of seconds.

Practical use is important if you want to get better at using your favorite tools.

3. Using A Chatbot

You can also create a chatbot to answer very specific questions by clearly specifying what role it should play in the instructions.

Here, I ask the chatbot to respond as if it were a doctor, while also using a touch of kindness and humor.

chatbot answerScreenshot by author, June 2022

AI-based chatbots can therefore provide personalized advice and recommendations based on customer preferences. Don’t hesitate to tailor the AI to respond in a particular manner.

A great example is that of Danny Richman, who created an AI version of Google’s John Mueller, called MuellerBot.

This bot builds on the above principle to answer SEO questions as if John Mueller himself were answering them.

It’s both fun and unsettling, as the answers can be quite accurate.

MuellerBotImage from Danny Richman, June 2022

4. Using An AI Assistant Program

Lastly, AI assistants for SEO are programs that run in the background and apply SEO fixes if a page is poorly constructed or has classic errors.

The first such applications date back to 2016, when Facebook was implementing bug auto-fixes with Getafix.

Based on all the bugs fixed in the past, the assistant prepares correction templates that are applied and reviewed by a human before being rolled out.

This is very applicable in SEO, where we know that concerns about meta tag titles, descriptions, pagination, and links are typical problems.

Facebook code baseImage from engineering.fb.com, June 2022

To do this, you can use GPT-3 in edit mode and modify the SEO pages using the appropriate instructions.

Below are my instructions:

  • Add a title with an H1 tag at the beginning of the text.
  • Add an <a href> link to the most important word in the body of the text.
  • Create useful outlinks at the end of text by using <ul><li>.
  • Add a YouTube video in the body of the text.
  • Put the top five concepts in bold.

modifying seo pages with AIScreenshot from OpenAI, June 2022

If we study the generated text, we can see that the results are excellent: The H1 title sums up the article, the words in bold are accurate, and the YouTube video and outbound links are relevant to the theme.

generated text using AIScreenshot from OpenAI, June 2022

In short: Your AI assistant can save you a lot of time.

Just a note: The links are all dummy links, but you can connect everything to a link database and use mapping tables to replace LINK1 with a link in your database or CSV file.

Now, you can appreciate the potential of automating these types of tasks.

Now that you know the different ways of interfacing a language model with your existing tools, don’t hesitate to implement the method(s) that works best for you, such as:

  • Chrome Plugin.
  • Directly into your CMS.
  • Via a chatbot.
  • Data visualization.
  • AI Assistant.

If you are interested in original article by Vincent Terrasi you can find it here