Last week, we got a lot of attention when we shared a previously unreported quote from Sam Altman—and we all need to figure out what the quote means, fast.
The quote came from a book called Our AI Journey by Adam Brotman and Andy Sack. In it, the two entrepreneurs interviewed Altman for the first chapter. When they asked him what AGI meant for consumer brand marketing, he replied:
“Oh, for that? It will mean that 95% of what marketers use agencies, strategists, and creative professionals for today will easily, nearly instantly and at almost no cost be handled by the AI — and the AI will likely be able to test the creative against real or synthetic customer focus groups for predicting results and optimizing. Again, all free, instant, and nearly perfect. Images, videos, campaign ideas? No problem.”
That quote made a lot of people in our audience sit up and pay attention. And rightly so. It implies a radically disruptive future coming to marketing and business very soon.
What does this mean for all our jobs? Our work? Our livelihoods?
I got some answers from Marketing AI Institute founder/CEO Paul Roetzer on Episode 87 of The Artificial Intelligence Show.
First, you have to realize this applies to everyone
“This quote can induce a bit of desperation,” says Roetzer. “What am I supposed to do next?”
If you feel that way, you’re not alone. We’re still trying to process it, too.
To start, you need to understand that Altman isn’t just talking about marketing. That’s the example he used in the quote.
“He’s talking about knowledge work,” says Roetzer. “You can substitute in here any knowledge work—accounting, sales, service, ops, HR, engineering, legal, all of it.”
No matter what your job is, if it involves knowledge work, this quote is relevant to your immediate future.
You can’t stick your head in the sand just because you’re not a marketer.
For Roetzer, the quote prompted him to game out where we are today with AI—and where we’re going. His analysis draws from 13 years of experience studying AI. He readily admits, like any forecast, he might get it wrong. But here’s where he thinks we’re going…
What to expect in 2024…
AI isn’t new. We’ve been researching the idea of giving machines human intelligence for 70 years.
For a long time, this research centered around machine learning, a core subset of AI. Machine learning made predictions about outcomes and behavior. For years, it’s been applied to things like pricing optimization and recommendation engines.
And that was valuable to businesses. But it wasn’t the AI we have now.
“What we have now started truly being kind of commercialized and accelerating development around 2011 and 2012,” says Roetzer.
This is when deep learning, the idea of giving machines vision and language understanding, really began to emerge.
After a decade of deep learning advancements, we have the ChatGPT wake-up call moment in late 2022.
Fast-forward to 2024. We are already getting more advanced large language models (LLMs). These models are now becoming multimodal. That means the models can understand not only language, but images, video, and audio. Google’s Gemini, for example, is built from the ground up to be multimodal.
The next frontier, says Roetzer, is reasoning.
“These models are going to make leaps forward in their reasoning ability,” he says. That means they’ll enable us to solve problems, plan, and make decisions even better than they do today.
Concurrently, we’re seeing far larger context windows. Google’s Gemini now has a whopping 1 million token context window. (And Google reports they achieved a 10 million token window in their research.) This is the amount of information the model is able to draw from to create the output.
Context windows, in some ways, lead to enhanced memory, says Roetzer. All major AI players are working to give models the ability to remember conversations and interactions. (OpenAI is already testing out a memory function for ChatGPT.)
This means a tool like ChatGPT would then draw from the context of your interactions when you use it. So, every person’s experience will be tailored to them specifically.
“You’re going to be able to set the parameters for yourself of how you want to interact with this thing, how you want it to talk to you, what tone, what style, what political beliefs, what religious beliefs, it’s all going to be personalized,” syas Roetzer.
At the same time, companies are hard at work making the models more reliable and accurate. So, you can actually start trusting them.
2024 is the year this begins to happen. We’re already expecting GPT-5 at some point this year. It will likely be accompaned with Gemini 2 and Llama 3 and other comparably powerful models.
“Multimodal reasoning, planning, decisioning, expanded context window, memory, personalization, reliability. That’s what GPT-5 class models will enable,” says Roetzer.
So what does this mean to businesses in the short term?
“I think it means we’ll start to see a scale up in adoption of AI,” says Roetzer.
He expects a rapid expansion of valuable use cases in business thanks to GPT-5 class models. This won’t happen overnight. But you are going to see a multi-year expansion starting soon.
It’s unclear right now how enterprises will respond. They may build on open models, close models, or a combination of both.
Increasingly powerful models aren’t just going to impact business in the next 1-2 years, says Roetzer.
“Scientific breakthroughs are around the corner.” Demis Hassabis at Google DeepMind, for one, believes we’ll see major breakthroughs in the next 1-2 years.
That’s because AI models can brute force a lot of data we already have that we cannot analyze and use at scale. So Roetzer expects breakthroughs from existing data in the near future.
Finally, in 2024, Roetzer expects us to start talking about AI’s impact on jobs. But not to the extent you might think.
“There will be stories of mass layoffs this year within certain industries, but I don’t think it’s going to be widespread. But you’re going to see some headlines,” he says.
He’s skeptical this is the year we see a mass impact on jobs. Though some new roles, like Chief AI Officer, will begin to emerge.
In fact, you might consider it the calm before the storm.
You’ll see every model get smarter. You’ll see every model get better at reasoning, planning, memory, and personalization.
But you’re unlikely to sit there and see a single inflection point where all of a sudden everything is different, says Roetzer.
Not yet.
What might be coming in 2025 and 2026…
Roetzer sees 2025 and 2026 as the years of multimodal AI explosion.
The models will have become much more capable in 2024. But now is when we truly empower them with all different types of data. For instance, over the last year, ChatGPT has learned how to process images. It cannot, as of writing, process video. That is coming. OpenAI’s Sora video model is a prelude to that.
It will take some time to teach each of the major models how to use all different types of mediums.
“I think we’re still 1-2 years off from where multimodal is just ubiquitous,” says Roetzer.
When that happens, true multimodal models will just work when you talk and interact with them through videos, images, and text.
In this phase, synthetic data becomes incredibly important.
Models learn from being given data. (ChatGPT was trained on the public internet.) To make them more capable, we need to feed them more data, faster. Some of these models are beginning to learn from observation. Tools like Sora will enable the creation of synthetic visual data, says Roetzer. This can then be used to quickly train AI models on vision tasks.
“I think synthetic data potentially becomes the dominant source of vision training,” he says.
This, in turn, leads to a further explosion of improvement in capabilities.
That, in turn, is when we start to feel their true power.
“Basically, these things become infinitely more valuable in a business environment and truly start to change the way we do work.”
The arrival of AI agents…
Roetzer estimates that somewhere between 2025 and 2027 is when we start to see an explosion of AI agents. AI agents are AI systems that can take actions for you.
We already have early examples of them today. “But I think this year is still just headlines, experimentations, and demonstrations,” says Roetzer.
He roughly estimates that AI agents are in their “GPT 1” phase at the moment. They still take lots of manual work to get to function properly. And when they do work, lots of oversight is required.
But, eventually, we’ll get to a point where they can take actions for you in a reliable way with little or no oversight.
Now, this is where we begin to get into the disruption that Altman talks about in his quote…