ImageFX is Google's new AI Image Generator

- Simplified AI -
Your gateway to discovering the world of Artificial Intelligence!

This week, Larry David, the creator of Seinfeld, found himself making an apology to Elmo on live TV…there's a sentence I never thought I would say or hear. Of course, it was just to promote the new season of his show "Curb Your Enthusiasm" on HBO.

Super Bowl Sunday is next week, and although I can't advocate for gambling, it's pretty funny to look at some of the weird prop bets associated with the game. You have your classic ones: which commercial will be shown first? What color will the Gatorade be that gets dumped on the winning coach? And this year there's a new one: how many times will the broadcast cut away to Taylor Swift during the game?

A sneak peek of what’s below:

• Where does AI training data come from?

• Google releases ImageFX, an AI image generator

• Arrow wants to help you nail the social media game

• Get creative: Mouthwatering pepperoni pizza!

What are web crawlers? How do they find the data that trains AI?

Turning AI topics into a walk in the park, not a mind-boggling maze

Just as gas and electricity are the fuel that powers your car, data serves as the fuel for AI Models. In past editions of this newsletter, we have frequently discussed how AI is trained on vast amounts of data. But where exactly does this data come from?

Although it's not near Halloween, let's talk about web crawlers. I'm not talking about spiders, of course, but computer bots that can scan the internet, acting as curators of information. These bots catalog information from websites, online repositories and other online sources like social media.

Some companies have developed proprietary web crawler programs, like Google's system for cataloging websites to enhance their main search functionality. But there are also open-source web crawlers, like Common Crawl, accessible to anyone with the right level of experience. OpenAI, the creator of ChatGPT, has disclosed using Common Crawl to gather data for training their early AI models.

Web crawlers are capable of searching the web for a variety of content, ranging from text to images, videos, and even social interactions. Social media sites, being the largest repositories of human interaction on the internet, provide a rich source of data. AI models utilize this data to learn how to mimic human behavior and understand how to interact with humans effectively.

The type of AI you're developing dictates the specific web data your web crawler targets. For instance, if you're creating an AI for building recipes, you'll concentrate on data from cooking and recipe websites. Or, if you're building a chatbot, then scraping human-to-human interactions from social media sites becomes crucial. A fundamental principle in training AI is the importance of the initial data gathered by web crawlers: "Garbage in = garbage out." AI learns by identifying patterns, and if the initial data is incorrect, skewed, or biased, the resulting tool will reflect those same flaws.

Why does this matter?

Web crawling is essential for making information accessible on the internet. Without it, Google, an integral part of our digital lives, would not exist—a scenario hard to imagine for many of us.

Despite its legality, web crawling for training data has attracted negative attention due to privacy concerns and copyright issues raised by individuals and companies.

Increasingly, large companies are implementing measures to counteract web crawlers. These defense mechanisms include:

• Blocking AI bots from crawling their websites to prevent cataloging of information.

• Updating their terms of service to prohibit the process, accompanied by threats of legal action.

• Developing counter-bots that obfuscate data on their sites, rendering it inaccurate for web crawlers. This produces "bad" data, potentially compromising an AI model if used in training without prior detection.

Large language models used for building general AI tools require enormous amounts of data for training - a necessity that mandates collecting this data from various sources. If the internet starts to restrict access to this data, it could put AI companies in between a rock and a hard place.

Now that you have an understanding of what web crawlers are and how they scan for information to feed into AI training, in next week's edition of Simplified AI, we'll explore the next step: how bots process information from the internet into usable data for AI development.

AI News Flash

Catch the latest AI news that is making all of the headlines this week!

Google's AI Image generator

First, we had Dall-E, followed by MidJourney, and now the newcomer, ImageFX, has arrived.

Google Labs' latest innovation, ImageFX, is set to redefine user interaction with AI image generators. Using simple text prompts and user-friendly "chips" (illustrated above in blue text), users can easily experiment with different versions of images.

Want to alter the position of the woman in the photo? Simply change the background "chip" to middleground or foreground.

Google said that initial experiments with ImageFX underscored the significance of fostering creative exploration, especially for individuals new to generative AI tools. Users frequently discovered innovative ideas by experimenting with a broad spectrum of prompts and concepts, refining their creations iteratively.

The image generation capabilities of ImageFX are driven by Imagen 2, a text-to-image model developed by Google DeepMind.

Honest Impression:

This tool falls short of the high expectations typically associated with a company like Google. After an hour of testing, it's clear that the images produced are average quality at best. Despite using "chips" labeled "ultra-high-quality," "4k," and "hyper-realistic," the results do not approach the quality offered by a tool like MidJourney.

However, the tool is user-friendly and fast. Images appear quicker than with other services, and experimenting with the "chips" to explore various combinations is pretty engaging. It's like playing ad-lib games, where the outcome can be vastly different from what was initially anticipated.

Why does this matter?

Ok, so the tool isn't the best, BUT I'm still sharing it for a few reasons:

• It's a free tool, whereas Midjourney requires a paid subscription

• It's beginner-friendly - the chips make it quick and easy to create new images

• Increased competition is always better, in my opinion - it encourages companies to build better features and tools

The standout feature in ImageFX is definitely its "chips", offering a much simpler, interactive method to build new images. This feature would definitely be beneficial to other programs, like Dall-E and Midjourney.

For more information on ImageFX, read Google's blog post, where they also highlight their MusicFX and TextFX tools.

Other Stories We're Following:

• Meta (Facebook) is looking to deploy its own AI GPUs to reduce dependency on other suppliers for building AI tools

• Amazon announces Rufus, the conversational AI that will help you become a better shopper

AI Company Spotlight

Explore trailblazing companies that are shaping our digital future

Arrow

"Automate social media posting for the cost of a large pizza" - that's the tagline you'll see when you first visit Arrow's website.

It's quite appealing, isn't it?

Anyone who has attempted to enhance their online presence via social media understands the challenges involved. Managing social media can feel like a full-time job, diverting attention from the core aspects of your business.

Arrow uses AI to help create curated social media posts for you and your business so you can have engaging content daily. However, it's not solely reliant on AI; they openly acknowledge the involvement of a "human in the loop," meaning your posts are reviewed by a social media expert before publication. While this might not seem as groundbreaking as a fully AI-driven service, the presence of this check-and-balance system is crucial. AI can occasionally produce errors, and having that human oversight provides confidence in the content being disseminated, particularly for professional businesses.

The feature I'm looking into right now is the ability to take a blog post or newsletter and turn it into digestible content for different social media platforms. It would be fantastic to use AI to automate the process of posting my newsletter and then having it create an Instagram, Twitter, Facebook, and LinkedIn post all at once.

I'm currently testing their tools out with a 14-day free trial and you can check out my thoughts and progress on twitter!

Get Creative with AI

Practice using AI and unleash creativity with your new collaborator

Copy the prompts below and paste them into ChatGPT, BingAI, Google Bard, or your generative AI tool of choice and let AI work its magic. Share your results with us…and take it a step beyond the initial prompts and see what you can create!

Text-based

• You've just discovered a hidden doorway in your closet that leads to a parallel universe. Describe your thrilling journey through this mysterious realm.

Visual-based*

• Generate a mouthwatering image of an irresistible pepperoni pizza emerging from a wood-fired oven, showcasing the perfect blend of crispy crust, bubbling cheese, and sizzling pepperoni.

Visit us on Instagram and Twitter to see how our examples turned out!

*Visual-based examples are for users of Dall-E, BingAI, or premium users of ChatGPT - may require paid subscription.