How can we better understand the current AI hype?

I enjoyed reading “AI as Normal Technology” by Arvind Narayanan and Sayash Kapoor recently. Their core point is that AI isn’t magic or menace; it’s a tool shaped by human decisions. Think electricity, not nuclear weapons.

I get why people worry about long-term AI risks. But a lot of that talk unfortunately feels like marketing and overhype. We’re also still fuzzy on the basics like what even is intelligence? It also echoes this really good paper, “Stop treating ‘AGI’ as the north-star goal of AI research”, co-signed by one of my colleagues, Margaret Mitchell.

The dystopia-vs-utopia framing doesn’t help. Narayanan and Kapoor argue for something more grounded: what’s happening in the research labs is not the reality of actual deployment, where diffusion is “limited by the speed of human, organizational, and institutional change.” It’s a reminder that with any tech, we tend to overestimate the short-term impact and underestimate the long-term shifts.

That hits close to something I’ve been thinking about non-stop: AI is driving this massive disintermediation wave, and in journalism, that means we urgently need to rethink how we build for audiences. It might be less about the tech, but more about driving change in newsrooms.

Are we taking AI seriously enough?

Surprisingly, I don’t think we are, at least not across the board. In many newsrooms, AI still feels like a geeky side project. That’s risky, because integrating AI in meaningful ways requires collective understanding, not just of what the tech can do, but what it can’t. And the best ideas for using it rarely come from a top-down directive; they emerge when people across teams understand the possibilities and start experimenting.

The most important question for the industry right now, in my opinion, is: how do we accelerate innovation and actually ship products that make a real impact? Not just cool experiments or one-off demos, but tools that genuinely change how we tell stories and serve our audiences.

That means building workflows and products with AI, not just reacting to it. Especially now, as tech companies move higher up the stack, shifting their focus from models to end-user experiences.

The good news is, the industry does have some solid fundamentals. As a former newsroom manager, I’ve seen just how tough digital transformation has been. The newsrooms that adapted best were the ones that broke down silos, built cross-functional teams, and made digital everyone’s job.

The biggest blocker isn’t the tech, it’s when the leadership doesn’t fully get it. From the outside, you can see the difference: some newsrooms are experimenting and shipping new things, others are stuck in neutral, even in how they cover AI.

Got any follow suggestions?

Andrej Karpathy is one of the rare AI researchers who can effortlessly shift between accessible, pedagogical videos for a general audience and deep dives into highly technical topics. His recent talk at Y Combinator is both fascinating and enlightening. It’s a masterclass in computer history and a powerful lens for understanding what’s happening in AI today. He coined terms like Software 3.0 and vibe-coding, framing how we’re all becoming coders simply by talking to machines in natural language.

He’s also refreshingly honest about the limitations. To him, LLMs are like “fallible people spirits”: they have superhuman knowledge but suffer from cognitive quirks like hallucinations and memory lapses. What resonated most with me was his vision for collaboration: rather than aiming to be replaced, we should focus on working with the machine, and creating efficient generation-verification loops. There’s a lot to reflect on here, especially for journalism.