Although there are lots of exciting new advances giving us unheard-of capabilities and resources with the output of large language models, there are some important casualties that we’ve recognized along the way.
Near the top of the list, and maybe at the top of the list, is newspapers. These paper media were institutions through the last two centuries, and to anybody who cares, it’s dispiriting to see what they’re like now.
Over the last decade or so, newsrooms kept getting slashed. But it’s not just head count. Journalists also became less able to speak truth to power through a number of trends, some involving liability.
This article from Politico says it vehemently, and in detail, talking about how writers end up in defensive stances and editors went from swashbuckling tigers to demure little cats. I think it’s highly relevant that the author used the word “swagger” a number of times to talk about the enormous change in tone, as well as content, over the years.
“Newspaper culture lost its conviction as it became aware of its own dimunition (sic),” Jack Shafer writes, under the scorching heading: Cultural Cache Wanes. “Yesterday’s journalists thought the world revolved around what their newspaper wrote. Today’s journalists resign themselves to the fact that their copy simply doesn’t matter as much.”
Ouch.
This is one of the more direct and frank pieces I could find on the details around how journalism has changed, and I think it should be mandatory reading for media undergrads.
ChatGPT is Speechless
Just for fun, I decided to ask GPT directly and simply:
“Why did newspapers die?”
The model thought for a while.
In fact, I thought it had frozen, and I had to restart the query, after which I looked at the little circle representing response latency, once again sitting there for at least a minute or so. That’s unusual, and lacking CoT, I wondered what the LLM was doing.
It finally spit out this:
“Newspapers declined primarily due to the rise of digital media, which fundamentally altered how people consume news. With internet accessibility, readers shifted to instant, free, and personalized content available online. Advertisers followed this migration, diverting revenue streams to platforms like Google and Facebook that offer targeted ads, undermining traditional newspapers’ financial viability. Additionally, social media empowered direct engagement and rapid dissemination of information, challenging newspapers’ slower publication cycles. Paywalls introduced to recapture revenue often discouraged readership, while cost-cutting measures reduced journalism quality, accelerating audience loss. The younger demographic, preferring multimedia formats over print, further eroded newspapers’ subscriber base. Lastly, mobile technology enhanced convenience, allowing users to access news instantly and conveniently, making printed newspapers less attractive. Together, these technological shifts, changing reader habits, financial pressures, and competition from digital platforms led to newspapers’ steady decline, marking the end of their dominance in the information landscape.”
Social media, pay walls, mobile technology, digital processes – all of it came together to overwhelm these businesses that just didn’t have the power to pivot in the right ways. Of course, the same goes for companies cranking out VHS cassettes, or film companies in the age of digital cameras. But newspapers are different somehow. It’s not just a product – it’s supposed to play a role, or arguably, several key roles, in our society.
And that’s not the whole story. Some of the more robust papers in older American cities were already on the ropes. But they hung on. Until AI became capable of actually writing that boilerplate copy that editors had been training their reporters to write. As David Brooks might say, they were then hoisted on their own petards.
A Journalist’s Story
In a recent TED talk, a writer about his own experience with ChatGPT, and taking a trip back memory lane, looking at what’s happened to newspapers.
Decades ago, he explained, he was a young reporter covering news in Connecticut. He talks about a boat ride that he took with a source, looking at how land around a power plant would be vulnerable to development through deregulation and the sale of utility land parcels.
At the time, he said, it was possible to pursue this story and try to let the public know what was happening.
Fast forward to today, and we really don’t have those capabilities anymore, at least not in the same ways. There’s no one to ring the bell, to respond to the bat-signal, and the tools that journalists and editors and newsrooms have used are largely ineffective now.
But in addition, Chesto recounts how in the digital age, ChatGPT doesn’t understand the story as it’s been archived for posterity.
He notes that when ChatGPT did a search on that old story, it came up with the erroneous idea that the land was not supposed to be used for hydroelectric power generation, when in fact, the opposite was true.
“Theoretically, you could over-develop and still pump water out of the lake and make electricity out of it, but Google didn’t understand any of that,” he added, explaining how the model also missed the context. It reminded me of some of an LLM’s other misunderstandings: neural nets can typically render a frisbee or a ball, but have trouble figuring out how those items move in real time.
The Biggest Factory in Massachusetts
Unsatiated, Chesto added another anecdote about ChatGPT mistakes – something we often call “hallucinations” or chalk up to the limited crunching of information related to a complex idea.
In this case, he mentioned how one of his colleagues got a hot lead that Dell employed 10,000 people at a facility in the state of Massachusetts.
Humans, he said, had to knock down that assertion, knowing that it was wrong.
“(The reporter) was paying for this service to find out where the biggest factory was in Massachusetts, and the robots got that one wrong, too,” he said.
So in a sense, the power of artificial intelligence has replaced manual, deliberate and painstaking journalism with collective digital searches that don’t always return the right result.
And that’s where we’re at.
Again, we’ve gained a lot with AI – but we’ve lost a lot too. It’s a big change. Unlike the cloud era and the big data era, AI is going to transform many aspects of our lives in big ways. We have to be ready to navigate this in the best way possible.