How Generative IA will Disrupt Everything in the Current Decade

Many will be surprised

Image by the author with Stable Diffusion

In recent months, AI systems like Midjourney, DALL-E, Stable Diffusion, LaMDA, and PaLM have made big strides in domains apparently as diverse as image and text generation. The capabilities of these systems are impressive: they produce highly suggestive images, create effective selling copy for advertising, and much, much more –all from mere “prompts” that describe what the user wants to get.

All this is done with Generative AI.

“Generative AI” refers to systems powered by deep neural networks that implement Large Language Models (LLM) in order to create some sort of content. Here I say “create,” meaning that it’s not a copy of something already existing, not in a philosophical sense (what is a “creation” anyway?).

Large new companies are emerging in this brave new world, like Jasper, which offers the generation of both selling copy and also images for advertising: Jasper now has a valuation of more than a billion dollars, becoming an overnight unicorn.

The first Generative AI platform to really make a dent was GPT-3 –released just a couple of years ago! After that, a succession of releases by several players in the field (OpenAI, Google, StableDiffusion, Google, DeepMind, and others) has appeared at a neck-breaking pace, so much so that it’s hard to stay current.

But beyond how fun and fantastic is to spend a while with Midjourney for creating images from our prompts, many tech enthusiasts struggle to make sense of this Generative IA wave.

Is Generative IA a solid trend, or is it just a fad?

I’ll go for “solid trend” because it will transform thousands of professional and leisure activities in the scope of this decade. Let me get started with an example.

I’m a massive tennis fan (at least in the TV sense). But live tennis matches take hours to finish, and I have other activities and interests, so I usually resort to watching replays or just highlights videos with the most entertaining 4 minutes or so from a match.

But what if instead of a 4-minutes video, I want 10 or 15 minutes one? Or if I want to include every point in the tie-breaks? I’m currently out of luck.

Now put your Generative-IA hat at work: a Generative IA sports video generator would create a video just for you according to the specifications that you informally put in a text prompt like the following:

Video of about 15 minutes with the most entertaining points of the Rafa Nadal vs. Tommy Paul match in Paris Bercy 2022, including complete tiebreaks if any, as well as every breakpoint converted

That’s it. You get a link with your personalized video, different from a video watched by anybody else in the world. And this video service would be as economically feasible as DALL-E and Midjourney.

Research is different from innovation. The former is concerned with published original results, and the latter has more to do with finding how to build a business from those results: innovation doesn’t care about originality but about growth, defensibility, investment return, etc.

Often things get confusing because research is done by companies like Google, which in principle are there to make a profit –but they understand that their business is high-tech, and tech isn’t high without research. So they get involved in financing research, as well as getting close to academia –many of their top researchers were hired from academia. As a researcher myself, I got invited to a Faculty Summit at their headquarters in Mountain View some years ago, and they lodged me in a suite at the Four Seasons hotel –whatever it takes to make a good impression on the academic community!

But even if it could be difficult –and even artificial– to make a clear cut between research and innovation, the difference is crucial here because, in the case of Generative AI, the two will be developed by different actors, and they will be associated with two different layers in the software stack –as pointed out by J. Currier:

  1. The bottom software layer is the Deep Learning model, built around implementations of Large Language Models (LLM) or equivalent internal representation. Models provide the base building block from which applications can be developed.
  2. The top software layer is the application one, which builds on top of the Deep Learning model to accomplish a specific task, for instance, to output an image from a text prompt.

This two-layer architecture will fuel a new era of accelerated innovation because once the bottom layer is developed by very large companies like Google, OpenAI, and others, smaller companies will provide the application layer –giving, of course, a cut of their profit to the bottom-layer provider.

Currently, the lower layer has been rapidly improved –and often, it has been distributed along with an application on top. For example, LaMDA and PaLM offer dialog capabilities out of the box, while DALL-E and Midjourney offer prompt-to-image services. But soon, the proliferation of open-source alternatives for the bottom layer will make it possible to develop just the top application layer and plug it into an already available bottom layer. Easier said than done, of course, but the fact is that the bottom layer is orders of magnitude more complex than the top one.

I’d argue that Generative IA will permeate almost every single knowledge work and leisure activity because it will provide tools for getting complexity away from formerly difficult activities and because it can provide a whole new level of personalization that I’d call “generative personalization.”

You can see what’s “generative personalization” from the sports video example above: each user is given a brand new and unique highlights video instead of just a selection between two or three options.

The cumulative impact from all the Generative IA applications is hard to exaggerate:

  1. Easy graphic creation is already within reach of non-professionals with tools like DALL-E, Midjourney, and Stable Diffusion, at least for simple utilitarian purposes like getting a header image for this post. Before this year, I was completely unable to draw my own images, and blog experts advised against wasting time on graphic design for your own stories.
  2. Photo editing users won’t need to endure a tough learning curve to master the intricate set of tools of Photoshop or Affinity Photo (I use the latter, and it’s so complex I have to consult YouTube tutorials to learn how to make most adjustments). With Generative AI, users will just ask the software to perform a given transformation, and voila! The image will get fixed. If Adobe fails to deliver Generative AI with their tools, they will be disrupted by new startups offering them and will go the way of Blockbuster.
  3. Presentation tools like PowerPoint, instead of just providing templates as they do now, will generate and fine-tune entire professional-level presentations from outline ideas. Currently, the difference between professional and amateur presentations is huge –this won’t be the case anymore.
  4. Text writing will be a process highly enhanced by Generative AI tools. Many forms of writing are already getting help from sophisticated tools like Grammarly, but Generative AI will give writers a qualitatively new level of help by, for instance, generating a complete first version of a blog. Writing will be a collaborative process between humans and the AI tool.
  5. Any software intended for a final user will have to be simple to use with text or voice prompts. User manuals and instructional videos will be a thing of the past, and as soon as users get used to the new simple way of using software, everything will have to offer it in order to remain relevant.
  6. Language learning will be done mainly with the help of voice assistants, which will be powered by –you guessed it right– Generative AI. Voice assistants, which will act like personal language coaches, will use their amazing natural language dialog capabilities, first seen in systems like Google’s LaMDA, to guide the human language learner in order to acquire vocabulary and expressions, improve pronunciation, etc. Language-teaching voice assistants is not a futuristic fantasy –it just makes economic sense as of right now.
  7. Even hardware products (like cars) will have Generative AI dialog-based help systems. Have you tried to perform a complex operation like adjusting the display in modern cars? Not easy, I can tell you. Instead of digging into complex manuals, you’ll just ask the voice assistant either to get instructions or directly get the adjustments done.

Many professions will be transformed beyond recognition. Graphic designers already feel the sting of this disruption. Entire professions will disappear, and other ones will be created. Powerful companies will go bankrupt, and new ones will become dominant, depending on how well they handle the tech disruption brought on by Generative AI.

And all of this will happen within this decade.

I may be wrong, but it seems to me that it was difficult, even for seasoned tech pundits, to forecast the enormous capabilities of the current image and text generators: it wasn’t evident a few years ago that huge models and training sets would lead to qualitatively different capabilities.

I’d go so far as to say that it was a fortunate, almost random finding. But now that we do have generative tools, the gates are open to innovating companies that will develop application after application at a fast pace: it’s mostly a matter of figuring out what can be radically improved and finding the suitable business model to make a business from a Generative IA idea.

A few years ago, it looked like other tech trends, like self-driving cars, VR, or blockchain, would soon take over, but self-driving technology has been limited by legislative hurdles, blockchain got hit by the economic downturn, and VR adoption is limited by hardware high costs. Generative AI, instead, is not yet limited by legislation (hey, polishing a PowerPoint presentation or generating a sports video is not a life or death matter) and doesn’t need expensive hardware to be bought by the user.

And we didn’t think that creative activities were going to be disrupted so soon. But they were.

We are entering new and sometimes weird times, where human creativity is mixed with machines’ new capabilities to the point that it’s hard to distinguish between them. As J. Currier points out:

“Today and for the next few years, this will feel surprising and in many ways scary. Because those creative moments where you go from zero-to-initial-ideas have always felt so uniquely human, because it has been so mysterious.”

How Generative IA will Disrupt Everything in the Current Decade Republished from Source https://towardsdatascience.com/how-generative-ia-will-disrupt-everything-in-the-current-decade-b4e8ce7dd4f1?source=rss—-7f60cf5620c9—4 via https://towardsdatascience.com/feed

<!–

–>

Time Stamp:

More from Blockchain Consultants