Jeffrey Funk does not share the widespread confidence that AI is already changing the world. At least, not the extent large language models has achieved so far.
Funk is an independent tech consultant and a former professor at Pennsylvania State University, Kobe University in Japan, and National University of Singapore, where he currently lives.
In his 40-year career, 25 of which were in academia, he has supported the growth of several Western companies and startups, and studied some of the most important trends in technology of the last three decades, such as the dot-com bubble and the rise of smartphones in Japan between the late 1990s and the early 2000s. His latest book criticises today’s venture capital investment bubbles in tech.
Jeffrey Funk’s most recent book, “Unicorns, Hype and Bubbles: A guide to spotting, avoiding and exploiting investment bubbles in tech,” was published last October.
“They’re telling us that everything is already changing right now, not in the next few years,” Funk told Mediatrends. “That’s what the increase in market capitalisation is about: big profit expectations for a lot of companies, thanks to the AI revolution. But is it?”
Still, every time an artificial general intelligence company launches a new model, it’s announced as it. were a watershed moment of the tech industry.
Proponents of AI will always talk about how their new models have done so well on their tests. But here’s the problem: they’re often trained on the very same tests they claim to excel at: business benchmarks, engineering exams. It’s self-reinforcing. And, frankly, meaningless.
A recent BBC research shows half AI assistants’ answers have “significant issues”, while several studies you’ve commented on LinkedIn indeed reveal a high degree of hallucinations in chatbot responses. Why this discrepancy?
First, because these are independent evaluations. Over the past six months, multiple analyses have consistently found a high percentage of hallucinations. And for many experts, that’s just how large language models work. They all rely on the same core technique: predicting the next word using probability. The result? Hallucinations are likely to persist.
For how long?
It’s difficult to predict, but hallucinations remain a major issue. While they might decrease over time, the decline won’t be immediate. If we apply the model of exponential technological change – something I’ve researched for years – it’s clear this won’t be solved overnight.
Image by Canva.
What’s a potential solution then?
Specialised LLMs, those trained on narrow, highly specific datasets, can help reduce hallucinations. Many companies are moving in this direction, refining AI on industry-specific data to improve accuracy. But several experts argue that a deeper shift in AI architecture is needed if businesses want true reliability. A new kind of AI will probably be required.
But companies and workers need AI now. Journalists, for instance. Newsrooms are increasingly employing it, though sometimes in controversial ways, as demonstrated by the Los Angeles Times‘ Insights tool.
Not all AI applications require absolute accuracy. For instance, AI-generated summaries or definitions don’t need to be flawless. Editors can use AI assistants to quickly gather a range of viewpoints on a topic, providing them a broader picture. Ultimately, it comes down to the journalist’s expertise. Experienced writers, who already have deep understanding of their subject, can use AI to polish and refine their drafts. Their experience helps them spot when the AI is wrong or when they need to adjust their queries to get more precise results.
Keep it simple, not asking complex human-like reasoning. Not yet, at least.
Yes, because if you’re trying to understand a controversy between two opinions, crossing them to reach a conclusion, then the resulting AI summary won’t probably give you the depth or nuance you’re looking for.
Time will tell how fast AGI improvements will unfold, but it will also determine who will lead the AI world, and whether China will surpass the US with new aces up its sleeve, such as DeepSeek and Manus. What will be the key factors in this race?
There’s a lot of uncertainty. AI is a vast field, with many researchers and organisations. New developments will emerge, but it’s crucial to remember that there’s a long gap between when ideas are first presented in academic papers and when they’re actually implemented. In AI, this timeline may not be as lengthy as in fields like physics, chemistry, or biology, but it’s still a long process.
Roughly five years, if considering the discovery of the transformer architecture in 2017 with the “Attention is all you need” paper and the launch of ChatGPT in 2022. What could happen in the next five years?
Something new will emerge, but in the meantime, the real issue is that the stock markets aren’t going to wait. They’ve been told, and convinced, that AI will add immense value now. The surge in market capitalisation is driven by the belief that AI is already changing the world, and many companies are expecting significant profits from it – not just AI suppliers, but also users of AI. However, so far, we haven’t seen any substantial evidence of this. So, while we can speculate about what might happen in the future, the simple fact that AI fails to deliver value to end users – the companies using it – could lead to a sharp drop in the stock market.
What’s in this AI story that is already working the way it is today?
AI for consumer services certainly isn’t one of them. These applications are struggling and haven’t been widely adopted. Instead, companies are leveraging specialised services and are seeing benefits from them, though they come at a significant cost.
The Berlaymont building in Brussels hosts the headquarters of the European Commission. Photo by Wikimedia Commons.
The efficiency issues affecting American AI companies could represent an opportunity for Europe to play its cards and secure a third seat in the US-China rivalry, or is it already too late?
I don’t think the competition is necessarily over for Europe. But it must be played on a different field: producing good applications. Applications that can significantly increase the productivity for the companies using them. American firms developing AI application software are facing uncertainties. None of these are generating substantial revenue. There is definitely a chance for European companies to come up with new alternatives.
There are some limitations, though. The first is a famous duality: the US innovates, the EU regulates. Does this reflect your opinion?
I tend to ignore this noise surrounding this argument. These kind of statements aren’t particularly helpful. The EU needs to focus on developing efficient, compliant applications for end users, and regulation plays a part in that, as citizens require safe tools. I see the bigger issue as the opposite: Europe is copying America too much.
Just a matter of seduction for the Silicon Valley model?
Unfortunately, this attitude reflects the broader tech world. People tend to copy each other. If someone comes out with a brilliant idea, most want to follow suit. When I was growing up, the semiconductor industry underwent numerous design and concept changes. This is the mentality the Internet was built on: thinking outside the box. And that’s exactly what Europe needs: to examine the problems their companies face when using applications to improve productivity and come up with innovative solutions. Instead, nowadays in the US, the venture capitals pour money into startups and founders, and make them billionaires far too soon, even before they have a product out there and have revenues.
Here comes the second problem: the disparity in investments, despite von der Leyen’s announcement of a new €200 billion AI plan. Leveraging open-source models to excel in specific tasks, as companies like Mistral AI are doing, offers a way to partially overcome the lack of funding. An emerging opportunity for European AI startups appears to be in military solutions.
I think European AI companies have significant opportunities across various sectors, including the defence industry. We’ve seen Ukraine’s innovative use of AI-powered drones to defend itself against Russian tanks and artillery. While Kyiv partnered with US startups, in this new, uncertain international landscape, where US support can no longer be taken for granted, Ukraine will have to come up with alternative solutions. The EU member states may step in and explore how AI can be used to defend its freedom as well. There’s an old saying in America: necessity is the mother of invention. Ukraine is fighting for its survival, and the EU wants to protect its freedom. Europe faces many critical challenges, and this is one of them.
