The excitement around the latest AI models, like OpenAI’s GPT-5, seems to be cooling off. Despite billions pouring into AI research and development, many experts are starting to question whether real progress is happening. The long-anticipated release of GPT-5 didn’t exactly blow people away. Instead, it’s been met with a muted response, and critics say the improvements are not as impressive as they might seem.

Are AI models really getting better?

Many in the AI community look at benchmarks and performance metrics to judge progress. GPT-5 does score higher on these tests, but experts argue those numbers don’t tell the whole story. Gary Marcus, a well-known neural scientist and critic of OpenAI, told The New Yorker that despite the higher scores, the new models aren’t necessarily more useful in real-world applications. In fact, he believes that for most practical purposes, these AI systems remain limited.

Since 2020, Marcus has advocated for focusing AI development on smaller, more targeted goals rather than trying to create a “general” AI that can do everything. His view is that the current approach, driven by large-scale data centers and massive investments, isn’t producing the breakthroughs many hoped for. Instead, it’s leading to bigger, more expensive models that don’t really solve the challenges that matter most.

The hype around scalable AI is fading

Big tech companies like OpenAI and Anthropic have been pushing what they call “scalable AI.” This means cramming as much data and processing power into these models as possible, hoping that bigger is better. The idea is that more resources will lead to exponential improvements. When OpenAI’s CEO Sam Altman talked about this in 2021, he believed that money could unlock near-human intelligence.

But reality has been less kind. Despite the huge investments, the pace of improvement appears to be slowing down. Experts are noticing that newer models are not significantly more useful than previous versions, even if they perform slightly better on benchmarks. This has led some, including Michael Rovatsos from the University of Edinburgh, to suggest that GPT-5 might mark a turning point—possibly signaling the end of trying to make ever more complicated, opaque models.

Are we approaching the limits of AI growth?

The idea that AI can keep evolving rapidly is now being challenged. A survey of nearly 500 AI researchers found that most believe achieving artificial general intelligence (AGI)—AI that matches human intelligence—is unlikely with the current development path. Bill Gates, the Microsoft co-founder, has previously said that AI development had hit a plateau.

Even the financial sector, which once bet heavily on AI’s promise, is starting to question the hype. For instance, CoreWeave, a major data center company that partners with OpenAI, saw its stock plunge 16 percent recently. This drop might be a sign that investor confidence in AI’s explosive growth is waning. The tech bubble that once surrounded AI seems to be deflating, leaving many to wonder what comes next.

While the future of AI remains uncertain, it’s clear that the era of unchecked optimism is coming to an end. Instead, the focus might shift toward more practical, achievable goals—ones that can deliver real benefits without the inflated promises of rapid, limitless growth. For now, the AI community is left to reconsider how best to move forward in a landscape that’s starting to look very different from what many once envisioned.

Inspired by

• https://futurism.com/scientists-worried-ai-pleateau

Sources

• gmpg.org
• newyorker.com
• technologyreview.com
• newscientist.com
• arxiv.org