404: Page Not Found. That error message has become a more common sight on government websites. Many — reportedly thousands — of federal government webpages were recently taken down, ranging from Census Bureau research on depression among LGBT adults to Food and Drug Administration guidance for making clinical trials more diverse.

These erasures come after President Donald Trump signed executive orders cracking down on diversity, equity and inclusion initiatives and what he calls gender ideology.

Marketplace’s Stephanie Hughes spoke with Jack Cushman, director of the Harvard Library Innovation Lab and a contributor to the End of Term archive project, which works to preserve government sites before a new administration takes over. They discussed his recent work archiving those sites and data sets and what’s lost when these digital artifacts are not properly archived.

The following is an edited transcript of their conversation.

Jack Cushman: I think one way to put it is that a great deal has been archived, and especially [by] the End of Term archive. Another way to put it is that a great deal has not been captured. We’re talking about a government of 2 million people who generate data in the course of doing their work, and whatever their work is, they’re going to report out to us as their ultimate employers. Here’s what I saw, here’s what I recorded, here’s how you can get access to a copy of it. So we’re talking about an infrastructure that is much larger than there’s any ability for the external community to make a copy of, so if we, in a serious way, if we shut off access to all of the data that the government has created for us, that we paid for, then we’re going to end up not having a copy of most of it at the end of the day. This isn’t something that we can fix from the outside, but it is something where we can get the copies of the things that are most important or most often used and help our patrons that way.

Stephanie Hughes: A federal judge recently ordered that at least some of these government websites be restored. Do you have a sense of how easy or how hard it is to do that?

Cushman: Sure. So I think it could be very easy to restore a website, or it could be very hard, depending how it was taken down in the first place. If you imagine that a website is a car driving along the road, and then you make the car stop, it could be, well, you just pulled over and you left the engine idling, and you just need to, like, hit the gas again. It could be that you turn off the keys. It could be that you threw the keys out the window. It could be that you removed the engine. And it’s very hard to tell in a given instance which one it is. I do think that if you shut down big running software projects and you let the people go who know how they work, and you let the running copies go and you’re trying to restore from backups, you can easily end up in a situation where it is expensive or prohibitive to get something running again.

Hughes: This taking down of web pages, it’s happening not just with the government. It’s happening with corporate websites. It’s happening with prestigious universities. What’s the cost of losing this information?

Cushman: So there’s a focus on government data right now, but the same issue applies to any other kind of information that you care about. So if there are videos that you remember seeing and caring about on YouTube, there’s probably only one account that controls those, and either a person or a Google policy could make that video vanish. If you have stories that you’re proud of publishing, especially if you’re publishing behind a paywall, there’s probably only one copy that most people access, and a change in policy or an accident or mistake could change the content of that forever, in a way that you would have trouble recovering from. The reason this happens is that when we’re working with digital materials, the incentive is to invest everything in the one best copy. So you get the one video-sharing website that everyone likes to use, and you invest more in that instead of having more copies. And when we get down to one copy of everything, we’re vulnerable to policy change, but we’re also vulnerable to accident and we’re vulnerable to cyberattacks, or we’re just vulnerable to losing our cultural memory. I think the impact of having a vulnerable cultural memory is that anything that we care about is harder to plan, and as we get on thinner and thinner ice, as we have fewer and fewer copies of each piece of data, there’s more and more vulnerability to just forgetting where we came from or important information that we had, and therefore making mistakes about what we do next.

Hughes: You said, economically, it makes sense to not have multiple copies of digital information. Can you tell me a little bit more about that? Like, it just actually costs money to have multiple copies of things on the internet?

Cushman: Sure. Imagine that you were a nonprofit funder, and you were trying to decide whether to give me money to make a new website that shares access to important public information. I could tell you, “I’m going to make a second copy of public information that everyone can already get from the government website directly.” Or I could tell you, “I’m going to make a copy of information that no one else has, that only we have at this library. That’s going to be the first time you can get it online.” Well, your funders are going to be a lot more likely to fund the story that is, I’m going to put information online for the first time. It’s also true at a business. If you’re asking for investment capital, if you say, “For the first time, people will be able to share their story,” that’s a lot more fundable than “I’m going to make a copy of YouTube,” even though YouTube already exists. No one wants to hear, “I’m going to make this for the fourth time.” They’re going to ask, “What’s better about yours than the one that’s currently the most popular?” And most of the time, the answer is, well, the one that’s most popular has the most money coming in, it has the most users, it has the most features. The current popular one is the best one, and we’re all going to go use that one instead.

So in the internet, you end up with this concentration where everyone is using the most popular version of the thing, and that’s fine as long as it works, and then if there’s any threat to the most popular thing, you don’t have that resiliency that you used to have. The other thing I like to compare this to is the issue of supply chains, and just-in-time supply chains that a lot of us learned that phrase for the first time in the [COVID] pandemic, when all of a sudden you couldn’t find toilet paper, and it turned out that our supply chains had a lot less redundancy or resiliency built into them. I think all of our data infrastructure online has that same just-in-time quality, where we’ve paid only the minimum amount we have to to keep it working today and not the amount to survive any kind of shock or challenge.

Hughes: I mean, I’ve had the experience as a reporter where I’ve saved a link to data that I was referencing in a story and then I’d gone back to it and it wasn’t there, and I felt like part of my brain was missing. What advice do you have for people who stumble across some data or some information on the web that they want to reference and hold on to, how can they do that?

Cushman: Well, the No. 1 advice for any kind of data preservation is lots of copies — keep stuff safe. And what that means for you is don’t trust remote things that you don’t control to be part of your memory. If you notice, oh, this is part of my memory, then say, I better figure out how to have a copy of this. And there are great tools to do that with. One of my favorites is something called Archive Webpage by the project Web Recorder, and that’ll give you either a browser extension in your browser or an application you can download where you can visit parts of the web and click Record, and everything that you see will be recorded into a file, which you can then save and keep wherever you want. I think the next step is to start thinking about how you can group together to invest in things like your local library or the local resources that you use, how to not rely on that one point of failure for everything that you do.

On a slightly lighter note, while working on this story, I wondered where the “404” error message came from, and I found a 2017 Wired story on its origins. It says that in the early days of the web, writing long messages took up valuable time and memory for coders, so a numerical designation was given to certain errors. 404 was assigned to “not found.”

I guess it’s more polite than the internet blowing you a giant raspberry, which is kinda what that 404 feels like when you come across it, followed by a sinking feeling in the pit of your stomach because the information you thought you could rely on isn’t all that reliable.

Venture capitalists have been welcomed into the Donald Trump administration, and their presence is growing. People who’ve been in the business of backing startups have been tapped to run the Office of Personnel Management and the Commodity Futures Trading Commission. Another, David Sacks, is the White House artificial intelligence and cryptocurrency czar.

Even the vice president, JD Vance, spent time making venture deals before he moved into politics.

Sarah Kunst, founder and managing director at Cleo Capital, says that in venture capital, you have to be good at saying no and comfortable taking risks knowing they likely won’t pan out. Marketplace’s Stephanie Hughes asked Kunst what it means to bring these qualities to the federal government.

The following is an edited transcript of their conversation.

Sarah Kunst: I think that that’s probably going to be a bit of a culture clash. The government and the people in America shouldn’t sort of be treated to a culture where it’s fine if 90%-plus fail as long as one or two does incredibly well. And regardless of political leanings, I do think that you ignore your base at your peril, and I don’t think that the average American wants to feel like it’s fine that they’re worse off, as long as one or two Americans are doing better than ever.

Stephanie Hughes: I want to ask just sort of about technology generally. You know, in Silicon Valley, there is this mindset of, why do with a person what you could do with technology? I don’t think they mean that meanly. It’s just kind of their jobs, to a certain extent. But if we have more venture capitalists working in the federal government, what do you think that will mean for the workforce and the amount of technology that we see?

Kunst: I mean, the idea that you always want people over technology is demonstrably not true. And we know that every time we get in a car and choose to drive somewhere versus walk. And so it’s not that all technology is bad and you always want people powering things. That being said, there needs to be a thought of, what are the checks and balances here? What are we giving up when we’re bringing in automation or when we’re bringing in technology? And then, more importantly, and I think this is the hard part for tech, is the technology ready to be responsible for the lives or the salaries or the well-being of hundreds of millions of our fellow Americans? That piece of it, I think, is the hardest for tech.

Way back in the day, far before Elon Musk, Twitter used to have this thing called a fail whale, which was when too many people would be on. [For example], it was the Super Bowl, it was whatever, and you would see this fail whale, which meant that the system was overloaded and it couldn’t work. And that is fine. It was sort of a meme. It was a startup. They were early days, you didn’t pay for it. Things happen, it’s OK if it breaks regularly. It is less OK if that same sort of build-fast, scale-fast, hope-it-all-works-out mentality, that works pretty well when you’re early-stage Twitter, starts to get applied to things like veterans benefits, where you can’t just give somebody a fail whale and say, “Sorry, you can’t get dialysis today and we don’t know what tomorrow is going to look like either because we’re doing something experimental, and that tech broke.”

And so I think that the balance here is going to be this sense of, how do you take being able to just sort of try things and if they don’t work, pivot, and iterate that into institutions where that doesn’t really serve the greater good, at least in the short term? And you’re often talking about real people’s money and lives. And so you can’t be quite as cavalier as you can about, “Oh no, the micro-messenger service that I use to tweet about my sandwich didn’t work for a couple hours today.”

Hughes: We’ve definitely seen the Silicon Valley ethos of “move fast and break things” in action in the last few weeks in Washington with the dismantling of whole agencies. How much more of that approach do you think we’ll see?

Kunst: I don’t think that they’re done. Elon Musk and his child in the Oval Office recently made it clear that he feels that he has the mandate and that there is a lot of dismantling to do. And I think different groups within the government right now have different feelings about how much needs to be dismantled, what needs to be dismantled and why. But there doesn’t seem to be much pushback inside the government on the executive level to the idea that more change is going to come. I think the question mark is does the Senate agree? Does Congress agree? Do the American people agree? And does the Supreme Court agree? But I think it’s clear to me, at least, that certainly the executive branch and a lot of the tech people who work in it now feel that they’re just getting started.

Hughes: You’ve been working in tech for a long time. You know a lot of people taking these jobs personally. What do you hope people who work in venture take the time to learn as they step into these new jobs?

Kunst: It’ll be interesting to see them learn how hard running a government is. You can’t sell a country, and you can’t have an IPO. There is no exit, you don’t wind it down, ideally. So you have to figure out this sort of perpetual motion machine of meeting the moment, meeting the technology. Just the sheer complexity of that, compared to in the tech world, where we tend to say, pick one thing and do it well. The government is not one thing. It can’t be one thing. It has to be everything. And so how do you even begin to do that? And so I think that a lot of their challenge is just going to be the sheer firehose that they’re drinking from that is the American experiment.

This story was produced by our colleagues at the BBC.

Lottie Hayton lost both her parents within two months of each other. As a young journalist she wanted to write about it, and in particular to investigate a new tech genre known as “grief tech,” or “ghostbots”.

An industry is emerging that uses artificial intelligence to build chatbots of people who’ve died with the aim of offering solace to those who’ve lost loved ones.

Hayton made a chatbot version of her dad and a visual talking avatar of her mom.

“My drive to try it was, I guess, in order to provide other people who might be using it with information,” said Hayton.

“The bot sort of blinks and moves slightly like she did, that was quite alarming. The face moves in a juddery way but it very much looked like her and that threw me off,” she added. “It had an air of her, but it was definitely robotic. I was inherently aware that this was a piece of technology.”

Within the last five years, the idea of digital resurrections have gone from science fiction to reality. Anyone can sign up to create a digital chat version of loved ones for as little as $10. Lifetime subscriptions, however, can cost several hundred dollars.

Justin Harrison set up his ghostbot business, You Only Virtual, after his own mother became ill.

“For anybody tech-savvy, it’s a pretty easy process, extracting text messages and emails and online messages,” Harrison said. “Our technology is nowhere near as good as it’s going to be in six months, in a year. Generally speaking it’s going to be mind-blowing in three years.”

This industry is still very new and still pretty niche. Businesses operating in this field only have a few thousand users and there’s not a huge amount of investor interest. But some experts believe that ghostbots could go mainstream.

Carl Orman is a Swedish researcher and author who has spent the past 10 years studying the ethics of the digital afterlife.

“Five years ago I would have said that most people would still find it kind of creepy. But then ChatGPT hit,” said Orman. “It’s not implausible that over the next decade or so, interacting with chatbots impersonating real humans becomes just as common as having a video call and that’s going to open up a new market for those chatbots. “

Experts are now calling for studies to find out if these tools can really help us with our grief, and if they do, figure out how companies might offer that ethically, consensually, and safely.

On this week’s Marketplace “Tech Bytes: Week in Review,” we’ll talk about Apple launching a new health research study and BuzzFeed starting a new social media platform. But first, the U.S. is pushing back against global AI regulation. This week there was a kind of who’s who of AI and government at the Artificial Intelligence Action Summit in Paris. French President Emmanuel Macron reportedly said there should be rules for this technology and that AI cannot be the Wild West. But the country that’s home to the original Wild West wants to forge ahead. U.S. Vice President JD Vance delivered a speech underlining the Donald Trump administration’s intent to develop AI without worrying about the risks. Marketplace’s Stephanie Hughes spoke with Jewel Burks Solomon, managing partner at the venture firm Collab Capital, about these topics for this week’s “Tech Bytes.”

There’s a concept in business called the first-mover advantage. Basically, it means that if you’re the first company with a successful product in a new market, you have the opportunity to dominate the market and fend off rivals.

But that advantage can be short-lived. Take Netscape, which produced Navigator, the first popular commercial web browser. Then Microsoft entered the field with Internet Explorer, and it wasn’t long before Navigator crashed.

In the world of AI chatbots, two of the first movers are OpenAI and Anthropic. But recently the Chinese company DeepSeek made a splash with an AI chatbot that it reportedly developed for a fraction of what its competitors have spent.

Marketplace’s Stephanie Hughes spoke with historian Margaret O’Mara, author of the book “The Code: Silicon Valley and the Remaking of America,” about whether America’s artificial intelligence industry should be worried about newcomers like DeepSeek.

The following is an edited transcript of their conversation.

Margaret O’Mara: There’s this dimension of foreign competition [that] also brings to mind the late ’70s and early ’80s in the U.S. semiconductor industry that was suddenly freshly challenged by advanced chips coming out of Japan, which was kind of a surprise. Japanese chipmakers were taking some technologies developed in the United States to develop complex chipmaking, and assisted by subsidies from the Japanese government, came to market and were able to rapidly undercut American chipmakers on price and really had Silicon Valley on the ropes for a few years in the early 1980s with this very fierce competition. So the DeepSeek saga brings to mind this earlier geopolitical moment, and I think there are some interesting similarities.

Stephanie Hughes: It’s way too early to make any pronouncements on where companies like OpenAI will land after this, but it does raise the specter of companies like Netscape and [social media pioneer] Friendster, who were sort of first, but certainly didn’t last forever. What factors could determine if American AI companies, you know, go the way of Friendster or if they can enjoy their first-mover advantage?

O’Mara: What I’m watching is, you know, how costly is it going to be to continue to develop these advanced models? And by cost, I don’t just mean cost of the consumer. I mean cost [of] energy, how efficient the hardware is. The capital expenditures of the largest tech platforms are mind-boggling, that kind of massive investment of capital and material is, you know, it’s not sustainable. It’s not sustainable from an environmental and energy-consumption standpoint, and also not kind of sustainable from a capital-expenditure standpoint. So I think that’s the real challenge, that it’s become so expensive to be a player in this. And there has to, what DeepSeek is pointing towards, is there is possibly another way. But this is different than before, than the age of Netscape or Friendster or other first movers, where the first movers in the AI space have, by necessity, had to access the capital of the large platforms. The large tech companies are the only ones that have the money and the resources and the data centers and all that data infrastructure to do these things, and that is something that is different than before.

Hughes: So actually, in preparing for this interview, I read this article from Harvard Business Review about first-mover advantage. It was older. It was from 2005 and one thing that made me really laugh as I was reading it is that it mentioned Apple as a company that had been a first mover and then, you know, had declined. And this was published two years before the first iPhone came out. And so I was wondering, is it possible to be, you know, a first mover and then lose that advantage and then start winning again? Like, is there a trajectory to it?

O’Mara: It is possible, usually by doing something different or the world has changed around you or something. Apple’s a great example of a company that was a superstar and then hit some roadblocks, so much so that by 1985, Steve Jobs is being fired by his board, as you know. “Get out of here! We need to do things differently.” And it’s only with the advent of the iPod and iTunes and then the iPhone that it really becomes the colossus it does. And that’s building something different. Every company has its ups and downs, but there’s this sort of really interesting sort of series of stories. It’s the company that becomes the verb in the beginning of a market doesn’t necessarily get to stay on top, and in some cases, they kind of fade altogether.

Hughes: What lessons can AI companies like OpenAI and Anthropic take from first movers, both the ultimate winners and the ultimate losers?

O’Mara: Well, I guess that maybe the top line is never take success for granted. As former Intel CEO, the legendary Andy Grove, once said, only the paranoid survive. So in a way, it’s reinforcing the hypercompetitive nature of the tech business, which is you got to keep on elbowing everyone out of the way. But also that, you know, the market you define in the beginning may change, and I think there’s a scale question. You know, like what happens when you’re really scaling up from being a startup that’s offering something completely new, helping define a market, and being able to take that to the next level, when the market actually becomes a mass market? What we also see is, you know, who’s going to deliver a product in the least expensive way, the smoothest way, the most frictionless way? Those are often the companies that, that come to dominate, but with the corollary [that] every firm eventually becomes a dinosaur, sometimes [it takes] longer than others, and it’s very difficult to stay, like, the innovator always. There’s a kind of a tension between, you know, being able to scale up and becoming a big market-dominant company and also continuing to be the one that’s developing the next, next big thing.

One thing about first movers: You can get attached to them or, at least, be nostalgic for what it was like to use them. I was talking with “Marketplace Tech” producer Daniel Shin about early movers we’ve used and loved, and he mentioned the search engine Ask Jeeves, which really brought me back.

Meanwhile, our senior producer, Daisy Palacios, remembers Facebook forerunner Myspace and obsessing over what song to put on her profile, though she doesn’t miss having to list her top eight friends, in order.

And Marketplace’s Meghan McCarty Carino, who had the idea for this segment, says given the alternatives now, she misses the original Friendster.

As for me, I never was on Friendster, but I do have a memory of being in college and my housemate Rob sending me an instant message over AOL Instant Messenger telling me to get on Friendster. You see, he had a classmate on there with more friends than he had, and he needed to catch up. I believe his message said, “She’s winning at life.”

Is there a first mover you wish was still around? Let us know at [email protected].