Content is King; AI is Joker.

7 min readMay 26, 2022

TL;DR: AI-assisted content generation capabilities are ramping up, enabling both professionals and amateurs to create much more. Already today AI can be used to generate high-quality content of various kinds, including natural language texts and conversations, visuals (image, video, 3D), speech, code, and more. Sooner than later, in many fields, AI will be involved in the creation of most content we consume and utilize, including architecture and internal designs, legal documents, entertainment content of various kinds, metaverse environments, travel plans, software testing scripts, protein designs, Ads, etc… Many professions are going to evolve (to say the least). The hype has reached the mainstream news and discussions (and it is real).

Prologue: A possible ultimate goal for AI is to continually understand and hold a universal representation of our world, then enable numerous add-on functionalities and applications such as search, classification, recommendation, exploration, generation, and more. We will achieve that by 2041 [20,41].
Meanwhile, there are already powerful latent representations for subsets of our world, including natural languages [2], natural images [3], programming languages [4], and combinations of the previous three [5, 6, 7, 8, 18], as well as protein structure [8] and many more.
With such powerful representations, new real or real-fake samples can be generated! There is significant research and progress in our ability to prompt Generative AI agents (/ neural-network models) to generate useful content. In this post, I will focus on content generation general use cases. You can further read about code-generation and its applications in my Software 3.0 post.

User Interface:

When planning an AI-assisted content generation UX/UI (user experience and user interface), these three aspects are to be decided upon:
1) interaction mode: copilot or automatic,
2) work unit (e.g. an image or a full album, document clause or a full document, code function or a micro-service, …),
3) starting point: updating existing content samples or inventing new content from scratch.

Let’s elaborate on the interaction mode options.
In Copilot mode, an AI assistant can, for example, suggest, auto-complete, extend, check, test, and improve the content. Usually done in iterations, guided by the user, and with small work units.
In Automatic mode, an AI assistant can, for example, i) replicate previous human actions or preferences and apply them to new samples, or ii) create or compose new samples with certain representation properties.

Which to choose? It depends, and should be considered per use case.

Use Cases Examples:

Creative:

Image creative:
Let’s start with the famous DALL-E [5] by OpenAI. DALL-E is an AI system that can create realistic-like images and art from a description in natural language.
The images below were created by providing a short text input to DALL-E:

Left: “The Thinker Samurai sculpture”, Center: “The Thinker Winnie the Pooh sculpture”, Right: “Photograph of humanoid robot lovers kissing in cyberpunk Kabukicho”.

And here are some more examples you can find on Twitter #dalle2:

Karen X. Cheng made a lovely video showing DALL-E in action:

Despite the awesomeness of the capabilities and results, the technology isn’t flawless. For example, in many cases, there are a lot of inaccuracies on the patch level. You can look at the photos above (or read more here [10]). Research is ongoing and growing. For example, Google Brain is also in the game (Imagen [1]), and here are some (of the fantastic) results:

Now, even Google’s tech isn’t perfect. Nevertheless, a good UX/UI could help mitigate some flaws. Additional commercial tools offering such capabilities, such as Bria.ai and NVIDIA Canvas:

Bria.ai promotion video

NVIDIA CANVAS

Video creative:
Automatic understanding and creation of video is more challenging than images due to the additional dimension of time. Yet, there are already a few relevant tools, for example those focusing on virtual or real human synthesis (most commonly talking-heads) scenarios: like synthesia.io and pantheonlab.ai. (Will we see a decrease in demand for human actresses and models?)

Pantheonlab.ai says this is a synthetic actress (i.e. the lady isn’t a real person)

A note on AI creativity:
People are very perceptive to visuals. So these capabilities presented above caught the public attention. Even main TV channels dedicated a story in the main news slot, and public discussions are all over twitter in all languages.
Thanks to technologies like the above, AI is now perceived to have creativity skills. These skills is largely related to the fact that the AI was trained on massive amount of data (created by creative humans). For example, for the “Chicken on a skateboard in Times Square” sample presented above, you may find multiple photos of a chicken on a skateboard online (surprisingly?).

Searching “chicken on skateboard” on google

Copywriting:

Grammarly and alike helps us fixing typos and grammar, as well as rephrasing based on desired sentiment. However, what if we need help with the creation of content. Gmail for example introduced an autocompletion feature. However, the autocompletion usually suggests a few words. What if we want full paragraphs, or for god sake even a blog outline?

New AI-assisted copywriting apps are flourishing, including jasper.ai and rytr.me. These can assist with long list of tasks, including: Blog Articles, Emails, Interview Questions, Landing Page Copies, Profile Bio, SEO Meta Description, Reply to Reviews & Messages, Story Plots, Ads Headlines, and many more.
I tried to create a Pitch for a startup idea:

From a one-liner business idea to three optional elevator pitches automatically created by Rytr.me. (Note: The above is my first attempt with the tool)

In my personal experience with these tools while trying to write this blog post you are now reading, the generated texts were helpful, in many cases, to unlock creativity and untie blockers.
However, the texts were inaccurate, inconsistent, illogical, and even broken in many cases, which caused a waste of time.
As you go, you learn to co-write with these tools and exploit their strengths while avoiding their weakness. In addition, different tools do have different strengths — for example, if your specific need is to rewrite sentences you can consider wordtune by AI21 Labs.

OpenAI GPT-3 and like Playground. I entered a beginning of a sentence and received back a few paragraphs (marked in green)

Ads, in scale:

Alibaba and Google [13] can auto-generate image and videos ads for you using combinations of your images, logos, and text assets.
With Alibaba’s Luban ([11], [12]) visual Ads generator, any seller can easily create ads for any of her.his products. Moreover, many different ads can be created for the same product targeting different audiences (!).

Code, tests, data:

Data is the fuel of the digital transformation.
Data generation could be useful in many ways. Here are a few examples:

¶ For testing and data science:

[15] Gretel.ai “APIs make it simple to generate anonymized and safe synthetic data so you can innovate faster and preserve privacy.”. They even state that they can provide “Synthetic data that’s as good, or even better than the data you have. Or don’t have.”
[14] Tonic.ai provides “Fake data that looks, feels, and behaves like production.”, “your data is modeled from your production data to help you tell an identical story in your testing environments”.
[16] Datagen.tech “Simulate the exact, high-variance, large-scale data your AI needs”. With Datagen you can “Accelerate your R&D and improve your computer vision models with access to a fully controllable synthetic data generation platform.”

¶ For code generation: See my Software 3.0 post

Future:

Content is King. And AI is the joker that can fill in the gaps.
AI-assisted content generation can empower individuals or form super-teams [17]. It can be used in co-pilot mode or automatically at scale. Any type of content can do, as long as there is a sufficient amount of data to train the AI.

In any case, the intent should be designed and controlled by humans.
Professions will evolve to focus on designing intent rather than working hard to create the content.

However, as we increase the usage of Generative AI, there is a lot of progress to be made in obtaining Responsible AI. Read a relevant and interesting post by Bria.ai CTO [19].

Last notes: I hear voices saying we are entering another AI Winter. I am actually surmising we are at the inflection point, and my mind is continuously busy thinking about it. I would love to hear what you think, whether in the comments or by contacting me on Linkedin.

Thanks!

And big thanks to my dear post reviewers: Avi Ben-Cohen, Ovadia Menadeva, Asaf Noy, Yair Adato.

References:

[1] https://gweb-research-imagen.appspot.com/
[2] Instruct GPT-3
[3] Scaling Vision Transformers
[4] https://openai.com/blog/openai-codex/
[5] DALL-E 2
[6] Flamingo: https://www.deepmind.com/blog/tackling-multiple-tasks-with-a-single-visual-language-model
[7] Gato: https://www.deepmind.com/publications/a-generalist-agent
[8] https://www.ai21.com/blog/jurassic-x-crossing-the-neuro-symbolic-chasm-with-the-mrkl-system
[9] AlphaFold, protein structure (Nature article)
[10] https://www.lesswrong.com/posts/r99tazGiLgzqFX7ka/playing-with-dall-e-2
[11]https://www.alibabacloud.com/blog/exploration-and-application-of-visual-production-technology_596929
[12] Automated Creative Optimization for E-Commerce Advertising
[13] https://support.google.com/google-ads/answer/9848688?hl=en
[14] Tonic
[15] Gretel.ai
[16] Datagen
[17] https://hbr.org/sponsored/2021/03/using-ai-to-turn-your-teams-into-superteams
[18] CLIP
[19] https://medium.com/@yairadato/do-you-understand-the-next-wave-of-ai-innovation-heres-what-you-need-to-know-fb05aaacba7f

[20] — https://www.ai2041.com
[41] — “The Game is Over!” 👇