Article contents:
- Introduction: The shaking of the AI throne
- Gary Marcus: A voice of reason in the noise
- GPT-5 disappointment: The moment everyone woke up
- Warning signs pile up: From Wall Street to MIT
- Lessons of history: Short-term pain, long-term gains
- The “gullibility gap”: Why no one listened
- Conclusion: Between magic and realism
Introduction: The shaking of the AI throne
It began with the underwhelming launch of GPT-5, followed by Sam Altman himself admitting to a “bubble” in market hype. It culminated in a comprehensive MIT study revealing that 95% of generative AI pilot projects fail. These events triggered widespread tech stock sell-offs, wiping $1 trillion from the S&P 500 and stoking fears that we are living through a new version of the dot-com bubble. Amid this turmoil, the words of a single critic who has warned of this scenario for years look particularly piercing.
Gary Marcus: A voice of reason in the noise
Since 2019, cognitive scientist and AI researcher Gary Marcus has warned of the limitations of large language models (LLMs) and the likelihood of a bubble. Marcus told Fortune: «I see myself as a realist and as someone who anticipated problems and was right about them». Marcus believes the massive hype surrounding AI blinded many to the technology’s fundamental problems, and that the market is now finally waking up to this bitter truth.
GPT-5 disappointment: The moment everyone woke up
Marcus attributes the recent market tremors above all to GPT-5’s “disappointing” performance. He explains: «It wasn’t a failure, but it fell short of expectations… GPT-5 was primarily marketed as artificial general intelligence (AGI), but it is not». He adds that this really “woke a lot of people up”. For Marcus, this comes as no surprise, having argued in 2022 that «deep learning is hitting a wall». Marcus feels we are like Wile E. Coyote in the cartoons running off a cliff and hanging in midair before falling: «That’s what it feels like. We’re off the cliff. This doesn’t make any sense».
Warning signs pile up: From Wall Street to MIT
Marcus’s warnings were not alone. In July, Torsten Sløk, chief economist at Apollo Global Management, warned that the top 10 companies in the S&P 500 today are more overvalued than they were in the 1990s bubble. He also pointed to massive spending on data centers to support theoretical future demand for AI. Then came a New York Times op-ed co-authored by former Google CEO Eric Schmidt arguing that «how fast AGI will be achieved remains uncertain».
Lessons of history: Short-term pain, long-term gains
Financial historians offer some perspective. They repeatedly document that this kind of frenzied investment typically leads to dramatic bubbles and crashes, but ultimately yields lasting value. Carlota Perez identified AI as the fifth technological revolution following this pattern, which eventually gave us railway infrastructure and personal computers. Therefore, even if the bubble bursts, a “golden age” for AI could follow.
The “gullibility gap”: Why no one listened
If Marcus was right, why didn’t people listen to him for years? It comes down to what he calls the “gullibility gap”. As a cognitive scientist, Marcus recognizes the human tendency toward “phenomenology”, attributing human traits to machines. He says: «People look at these machines and make the mistake of attributing an intelligence that isn’t really there… This entire market was built on the misunderstanding of people». These machines may look human, but “they don’t actually work like you”, Marcus says, adding: «It’s almost tragic».
Conclusion: Between magic and realism
The market finally seems to be distinguishing between magic and reality. AI may be a transformative technology, but it is not a silver bullet for all problems, and the path to realizing its full potential will be longer and more complex than many thought. And while some view AI as “like magic”, the reality is that it has not changed the world quite yet. The current backlash may be precisely the reality check the industry needs to move forward on a more sustainable footing.
Frequently asked questions
Q: Who is Gary Marcus?
A: He is a cognitive scientist, researcher, and AI entrepreneur known for his long-standing criticisms of the mainstream “deep learning” approach and warnings against overestimating the capabilities of large language models.
Q: What is the main problem with GPT-5 according to the article?
A: The problem is not that it is a bad model, but that it was “disappointing” and failed to represent the quantum leap many expected, sparking doubts about the sustainability of rapid AI progress.
Q: Does this mean AI was just a bubble?
AI is not necessarily just a bubble. The article points to a historical pattern where major technological revolutions lead to financial bubbles that burst, yet leave behind infrastructure and lasting value. This may simply be a correction phase before a “golden age”.
