Why AI can never replace the news

ChatGPT’s worldview and its responses have become a new node in reality, much like how mainstream media outlets are nodes in our common information weave. Some already believe that AI will wipe out news organizations altogether.

The written word has been challenged repeatedly throughout the 20th century, and its demise has been predicted over and over. First, radio was supposed to kill the newspaper, then TV, then the internet. Now it’s AI.

Yet if news outlets disappear, no one will know what the news actually is. AI services rely on newsrooms to function.

“Citizen journalism,” which Elon Musk and other tech bros put their faith in, has never worked. It turns into garbage, anecdotes, and propaganda. Real journalism costs money to produce. It’s a craft that takes time to master.

But as you probably know, AI apps can now also browse and search the web for information.

What about the buzzword “Agent”?

According to the PR departments of AI companies, the “agent stage” is in development. Or, it was, last year. Do you remember that we were going to pay 2,000 USD per month, per agent? Well, according to Sam Altman and OpenAI we were.

Now, almost nobody is talking about those agents. The idea was that your AI could act on your behalf more independently, continuously running in the background. But that did not work. The “agents” made mistakes, just like the rest of the LLM AIs. Otherwise, OpenAI would use themselves to make more money.

The usual example is something like ticket sales. You should use your AI to buy tickets, is what AI companies seem to suggest. Your AI agent will search for flight tickets.

But that’s just misguided and mostly hype, since ticket vendors are already incorporating AI algorithms to make things easier in various ways. You don’t need your own AI to buy tickets. There will be plenty of so called AIs offering those services.

But why not let an agent buy and sell stocks according to your own strategy, around the clock? What could possibly go wrong?

Joking aside: the word “agent” is unnecessary. It’s just buzzword marketing. The goal is still to try to get an AI to do something useful for you. The “agent” would dive deeper, think in parallel, and deliver something more comprehensive and connected. Well, that was the idea. But it’s still the same LLM—the same hallucinating AI—possibly with slightly fewer errors. But still with errors. If one step in the agent-process goes wrong, the whole thing turns sour.

It’s about attention

AI companies are competing with traditional web companies. FAANG (Facebook, Amazon, Apple, Netflix, and Google) led the previous race for viewers’ attention. The new race is among AI providers. That’s why Elon Musk has launched his own AI platform. He’s caught up in the hype, thinking that AI has greater potential than even rockets or cars.

So, what does the lineup look like and where are they located? Here is the full list of the current landscape of competitors, in alphabetical order: Amazon Q, Amazon, USA, ChatGPT, OpenAI, USA, Copilot, Microsoft, USA, Claude, Anthropic, France, DeepSeek, High-Flyer, China, Gemini, Alphabet (Google), USA, Grok, xAI, USA, Meta AI, Meta (Facebook), USA.

Six out of the eight are American.

Challenging them would cost billions of dollars. We’ll see if IBM and Apple respectively step up with what’s needed.

Geopolitical AI

Could more players enter the field? Yes, it’s reasonable to guess that both the EU and India are large and democratic enough to launch their own AI platforms. In that case, we’d have 10 public-facing platforms to choose from.

There are more platforms than the ones listed above out there, such as those in the military, but they serve a different purpose. The AI companies listed above target the general public and might be the new “FAANG” group if AI development keeps progressing as it has. “AI8” could be a fitting abbreviation?

But you can’t buy DeepSeek on the stock market, unlike the old FAANG companies, which are all available on U.S. exchanges. Only three FAANG companies overlap with “AI8”: Meta (Facebook), Alphabet (Google), and Amazon. Maybe Netflix can make a comeback once AI can generate full-feature-length movies?

Perplexity is another company that’s gotten a lot of attention, even though they don’t develop their own AI engine. Instead, they adapt others’ engines for their search service, allowing Perplexity users to switch between multiple AI platforms.

As a user, you could benefit from this, at least from a privacy standpoint, since your data is stored with Perplexity rather than with Microsoft or X. But is one really safer than the other?

AI is still growing

ChatGPT is still growing rapidly in 2025, even though Google has managed to slow the churn from search to chat by introducing its own AI summaries. Still, it seems like ChatGPT is the big winner compared to the rest. The first-mover advantage is strong, and the letters “ChatGPT” has become synonymous with LLM’s, kinda like the word “Jacuzzi” is used for most hot bath tubs, whatever the brand.

For instance, Elon Musk’s challenger X hasn’t made much of an impression lately. There’s nothing functionally that really stands out about Grok compared to the other platforms.

What we’re seeing instead is that an AI service’s unique traits only make headlines at launch. Grok was a bit more sarcastic than the others, which generated some buzz. Others prefer ChatGPT’s “personalities.”

All the AI models above employ censorship in various ways. Some try to get DeepSeek to comment on sensitive political topics. DeepSeek refuses to answer such questions about China.

It’s pretty clear that no AI today can compete without modern chips from Nvidia, even if DeepSeek claims to use them more efficiently than OpenAI does.

I asked DeepSeek about how the service has developed. DeepSeek itself replied that its parent company acquired more than 10,000 Nvidia GPUs before the sanctions were introduced. Lately China has been turning away from imported processing chips, working on developing their own lines of GPU’s.

Rumor has it that they now have over 50,000 such GPUs—perhaps significantly more.

And if you just have good enough AI algorithms, you win. TikTok is a prime example.

AI can still disrupt

In other words, whoever develops the best social media algorithm has a shot at wiping out the competition. Much like how Google dominated search for decades. Now that more and more people are leaving Twitter/X, a new AI could step in and take over that role.

Personally, I don’t think all of today’s players will survive. The revenues do not match the expenditures. Elon Musk’s X is fragmented in many ways. Google hasn’t been very responsive over the years and hasn’t built much user loyalty. They never responded to criticism. They actually arrogantly dismissed users’ wishes for decades, because they didn’t have to do anything. For example; they’ve always neglected to provide decent support. And they kept those paid links at the top of the search page for as many years as they could, never developing the search results.

In short, they did everything to maximize profits while doing as little as possible for the users. That’s how powerful the PageRank invention was—it fended off all challengers for decades. Now they’re trying to regain growth, much of which has shifted to OpenAI.

Social AI?

These days, AI platforms are better and faster at answering questions than the old Google ever was. And it’s high time Google faces real competition. And thanks to this competition from OpenAI, Alphabet can now keep its Chrome browser, as Cryptopolitan reported on recently.

Now, the question remains if the current crop of LLM (Large Language Model) AIs can turn a profit in the future. New technologies are often overhyped at the start, but underestimated in the long run. Whoever first integrates a well-functioning social network into their AI might dominate the entire market for the coming decades.

Already we see a lot of signs that companionship is a possible profitable avenue for AI companies. To offer therapy, or a friendly chat partner, like in the movie Her. That could lead to even lonelier individuals, and even psychosis.

I also note that no one in the news industry recommend AI as a platform for news delivery.

Only people outside the field do that. They obviously have not tested whether the “news” AI provides is actually news at all.

Every time I test the various AI platforms on news topics, the result is the same: no AI can distinguish today’s news from yesterday’s, or from six months or six years ago.

News articles are too similar over the years. Almost the same events repeat, which confuses the AI networks. There are exceptions of course, if the news events are very simple and unique, and your question is tailored to that event. At times the AIs manage to get their replies correct, but sometimes not. That’s not a good basis for news reporting.

Journalists can usually keep track of timelines and who’s who. AI platforms, however, often serve up old, confusing, and outright erroneous news, often omitting important parts of the topic.

But you only notice if you double-check the information your AI is providing.

But, you interject; the LLM (Large Language Model) AI engines will become better, so they will be able to tell what is news in the future?

No, due to the nature of an LLM. It’s a hallucinating machine to its core. An LLM doesn’t know what reality is. It cannot successfully differentiate between fantasy and the real world, between PR, marketing, lies, propaganda and misinformation. The LLM’s will continue to hallucinate.

So, is it possible to replace traditional news sources with an AI delivering updates? The answer is clearly: No. Because; answer this: what’s an acceptable error rate for a news service, in your view?