Beyond "Breaking News": The State of the AI Union

Overwhelm.

That’s what I hear most often from people about AI.

“I feel overwhelmed.”

And guess what? So do I.

I’ve been on a lot of airplanes lately and there is something about flying that always gives me perspective. As I look down today from 30,000 feet on the contiguous 48 United States and think about the year that’s entering its final weeks, I have one thing to say to you:

It’s time to stop the Breaking News cycle.

Now, don’t get me wrong – I’m clearly an AI enthusiast, eager to be on the cutting edge of the latest developments. But the constant stream of alarm bells, notifications, and race to be the first “in the know” is not serving any of us well. It’s hindering adoption by making it seem like:

• AI is moving so fast, how do I keep up?
• There are so many tools, how do I pick the right one?
• What about my data, my privacy - who can I trust?

I give a lot of thought to what I serve up in this newsletter that will be useful, not just add more noise. And from what I’m seeing and hearing, it’s not the latest tricks, cheat sheets or strategy documents you want.

It’s a moment to think, reflect, and prepare.

So, for the rest of this calendar year, we’re going to take it slow. We’ll look back at where we were a year ago, take stock of where we are today, and stand firmly grounded for the future.

This is the State of the AI Union.

In this series we’ll review at a high level:

• Media Generators (Image, Voice, Video, etc)
• Large Language Models
• Enterprise Applications
• Product Development

I’ll focus on the underlying technology, leaders in each category, and ethics, as always.

But first, let me start by trying to answer: Why does it seem like AI is moving like a bullet train through our living rooms?

The Illusion of Modern Media

The breakneck pace of AI development is an illusion driven by modern media. To feed the constant cycle of clickbait “breaking news,” most AI-first companies are releasing technology in beta form, using customers as free labor to train models. Capabilities are unveiled piecemeal as individual features and fragments rather than full products.

We are in a post-product world. There is not an original, mature, robust generative AI product on the market today. Not one.

Instead, we see a new phenomenon emerge: the AI playground. Cloud computing enables real-time delivery of incremental capabilities. Users serve as unpaid testers to improve these half-baked initial releases. The dominance of short form, snackable content shapes perceptions where each new feature seems revolutionary yet instantly outdated.

In this environment of fragmented progress, it's no wonder people feel overwhelmed by the pace of change. But to truly evaluate the state of AI, we need to step back from the manufactured hype cycle. Examining the underlying forces reveals an industry incentivized to reward constant, premature capability releases over thoughtful innovation.

Cloud Computing

The rise of cloud-based delivery models has fundamentally changed how AI capabilities are built and released. In the past, software was painstakingly developed, heavily tested and vetted, then shipped out to customers as a "completed" product.? This was true historically for Microsoft Excel, Adobe Photoshop, and other iconic applications.

Today, that cycle no longer applies. Cloud infrastructure allows companies to push out incremental updates, new features, and partial capabilities in real-time. There is no need to wait until everything is "done." Instead, barebones beta releases can be deployed quickly based on what's ready at the moment.

We see this clearly with new AI features. Take "generative image inpainting," introduced by OpenAI in DALL-E2 back in 2022 . This allowed users to remove an object from an image and replace it with something new. A huge capability at the time! Yet today, it's a standard feature incorporated into every new image generator. Within one year it went from headline innovation to table stakes.

Take Microsoft Excel as if it were released today. Imagine if every minor capability was trumpeted as a breakthrough: “Breaking News: Adding a column of numbers just got easier with Microsoft’s groundbreaking summation function!” This drips out functionality that would have previously been bundled into major releases. It sounds ludicrous.

For users, this creates a perception of constant change and innovation. New features roll out at a dizzying pace. The risk is Core capabilities may remain unreliable or unfinished, as teams spread themselves thin racing to pump out a stream of microscopic advances. For companies, there's incentive to operate products as "perpetual betas," buying time to perfect the underlying foundation.

So while cloud delivery models enable unprecedented agility, they also risk encouraging fragmentation over focus. We see this as some of the most technically capable players like Google, who should be leading AI advances based upon their deep investment in the field, fumbling as they try to keep up in this feature focused environment. For AI systems especially, incomplete releases burden users and create an illusion of revolution over iteration.

User Playgrounds

Self-service platforms are becoming ubiquitous across many industries, from travel booking to retail. Users are now expected to serve themselves without service agents. We've come to assume this exchange without any reduction in the price of the goods and services we are consuming.

In the past, most software required extensive training and expertise. Products like Photoshop, SAP, or AutoCAD catered to professional power users. But AI has helped enable a new wave of "self-service" platforms - tools like ChatGPT that are intuitive and accessible to everyday users.

This self-service model relies on users themselves to refine and improve the systems. Developers release MVP versions earlier, using the public as free QA testers. With constant real-world usage, the AI gets rapid feedback to learn and enhance its capabilities.

This effectively turns the users into unpaid trainers and validators. The prototyping and testing that used to happen internally now happens in public view. Every query and interaction represents a training signal to the models.

This creates a kind of "user playground" effect. Individuals get to experiment and derive value, while providers benefit from crowdsourced improvements. It feels organic and community-driven, but represents a major shift of labor onto consumers.

Snackable Content

Digital media has conditioned us to have ever-shorter attention spans. On platforms like TikTok and Instagram, content is condensed into bite-sized snippets lasting seconds. Even written content trends toward scannable listicles and quick takes.

This drive for ultra-concise information shapes how we engage with new technologies like AI. Capabilities are unveiled piecemeal and most demonstrations focus on the quick "wow" factor rather than substantive evaluation.

We see AI products emphasized for their flashiest functions - image generators pumping out novelty art on command, chatbots delivering clever quips. The trend is toward immediately impressive yet superficial capabilities over deeper utility.

Of course, these attention-grabbing demos attract interest and uptake. But snackable content risks trivializing complex technologies. When something new pops up daily, it overwhelms our ability to fully assess impact and ethics.

So while snackable content makes AI more consumable, it also fosters a mentality of constant novelty-seeking. The drive to stay relevant risks rushing capabilities out as rapidly-consumed morsels rather than meaningful innovation that users can deeply integrate.

Talk to Me Like an LLM

Amidst the breaking news, researchers are systematically and methodically studying the very technologies that the world is beta testing, and the lessons are so undeniably human.? According the finest computer scientists, it turns out that telling chaos machines to do the following four things helps improve its performance.? And if it works for machines, perhaps it's good advice for humans as well: ?

Take a deep breath. The pace of AI development might seem overwhelming, but it's important to remember that much of this feeling is driven by media hype. The actual evolution of AI is more measured and manageable. Approaching AI with calmness and clarity allows us to better understand and integrate these technologies into our lives.

Work on this step by step. Embracing AI doesn't mean getting swept up in every new trend. Instead, focus on understanding each development and its practical application. Just as we guide LLMs with structured prompts, we can methodically explore AI's capabilities, ensuring we adopt technologies based on their true value, not just their novelty.

Believe in your abilities to engage with AI. Despite the complexity, each of us has the capacity to learn and adapt. Confidence in our ability to understand these technologies can transform overwhelm into empowerment.

Stay determined. The journey through AI's landscape is ongoing and evolving. Perseverance in learning and adapting will yield a deeper understanding and capabilities over time.

I’ll see you same time, same place, next week…without breaking news.

I'm Lori Mazor and I teach AI with a human touch, empowering intelligent business. If you enjoy this newsletter, here are more ways we can engage:

• Join me for the launch of TEMPERATURE: Creativity in the Age of AI
• Register for the Generative AI Bootcamp
• Invite me to speak at your company event or lead a training session