At Davos, AI Is Really The Next Big Thing

"Davos, Switzerland -- I tried to count the number of times "AI" appears on the signs, walls and banners along the main Promenade here at Davos. I lost count at 67. AI, AI, AI -- it was the topic at the year's most high-profile gathering of global leaders and thinkers. (OK, fine, they also talked Trump and tariffs.)

What AI progress is coming and how will it impact our jobs, work and more? Last week, I had dozens of meetings with tech executives, analysts and academics, some on stage at our Journal House.

Here's what they were all buzzing about -- and my take on it.

Agents inbound

Let's be clear, no one here really agreed on the definition of "AI agent."

But for now, let's define it as an assistant that can take real-world actions on your behalf -- send emails, book flights, shop and more.

Kevin Weil, OpenAI chief product officer, described the new ChatGPT Operator feature that can handle tasks in your browser, like filling out a soccer registration form for his kids.

Meanwhile, Anthropic CEO Dario Amodei said a Claude-powered tool capable of taking multistep tasks on our behalf -- like writing code, testing code, talking to coworkers about it -- could launch in the first half of the year.

My take: I'm excited to test these, but let's not forget how often AI still gets things wrong. Do I really trust it to handle irreversible decisions? Weil reassured me that ChatGPT's agent will check in for confirmation before acting -- we'll see how that works in practice.

The bigger conversation here at Davos was how AI agents will reshape work. If AI takes over these tasks, what do we work on?

There's a lot of talk about people becoming "agent managers" -- which sounds futuristic, but mostly makes me wonder how long before I have to manage an AI that makes dumb mistakes.

Smarter models

The big AI players -- Anthropic, Google, Meta, OpenAI -- are in an all-out race to release faster, smarter and more capable models.

There's also a new focus on "reasoning" models, like GPT-o1 and o3, that don't just generate responses but take time to think through them to deliver more accurate, nuanced answers.

On Thursday, OpenAI released o3-mini. Weil says the company is already training its successor.

As for Anthropic, Amodei also teased future models and said they'd be released in the not-so-distant future. When I asked him about the industry's growing obsession with "reasoning" models, he said he doesn't see them as a separate category. Reasoning is already built into existing models like Claude Sonnet 3.5 rather than being some entirely new breakthrough, he said.

My take: You don't have to be a full-blown AI nerd, obsessing over benchmarks and parameter counts, to notice these models are getting better. In fact, I've now reluctantly accepted that I need to pay for both Claude and ChatGPT because one excels at some things while the other shines elsewhere.

Human-level intelligence

Every major AI company is chasing human-level smarts, often called Artificial General Intelligence (AGI). But when will it happen?

Amodei on Tuesday: It will be about two to three years until "AI systems are better than humans at almost everything. . .then eventually better than all humans at everything."

Weil on Wednesday: "I don't even know if it'll be 2027, I think it could be earlier."

Meta Chief AI Scientist Yann LeCun on Thursday: "That is all complete BS." LeCun believes large language models won't get us there and we need new systems

My take: It's hard to know who to believe here. Do we believe the AI executives who have insight into what's happening in their research labs but also have incentive to hype this technology to investors, consumers and beyond? Do we believe others who say there are technical limitations? Either way, I highly doubt that in two years time, machines will be better than humans at everything.

I mean, sure, AI, please go ahead and try to fold a fitted sheet. Or install a car seat in a car. I'll wait.

Autonomous vehicles

It wasn't just generative AI -- autonomous driving was also a hot topic among tech leaders.

The job of human Uber drivers will exist for the following decade, Uber CEO Dara Khosrowshahi told me in an interview.

"For the next 10 years, you are going to have hybrid networks of humans and machines," he said.

In Atlanta and Austin, Texas, you'll soon be able to hail a self-driving Waymo via the Uber app.

Uber's committed to being the on-demand platform for transportation, groceries, takeout and more -- essentially beating Amazon at same-day delivery.

Whether that's done via humans driving cars, autonomous sidewalk robots, autonomous cars or drones, Khosrowshahi doesn't seem to care." [1]

Driving humans are so expensive and dangerous that they will disappear completely as class very soon. Khosrowshahi still temporarily needs them so he is saying BS just to calm them down when they are driven for slaughter.

1.  At Davos, AI Is Really The Next Big Thing. Stern, Joanna.  Wall Street Journal, Eastern edition; New York, N.Y.. 27 Jan 2025: A11.