For years, artificial intelligence was the darling of Silicon Valley and boardrooms worldwide—a technological marvel promising to revolutionize industries, upend the way we live, and usher in an era of unprecedented productivity. Yet, as summer 2025 draws to a close, a wave of skepticism is washing over the AI sector. The once-unquestioned march of progress has slowed, and the industry now faces tough questions about its future, its limits, and its real impact on the workplace and society at large.
According to NPR, the mood shifted sharply following several high-profile developments in August. MIT released a report on August 23, 2025, that sent shockwaves through the business community: a staggering 95% of generative AI pilot projects at companies are failing. This sobering statistic landed just as consumers expressed disappointment with the latest version of ChatGPT, which had been released earlier in the month. The sense of unease was compounded when OpenAI CEO Sam Altman publicly floated the idea of an AI bubble, a comment that contributed to a notable dip in tech stocks.
To unpack these developments, NPR’s Scott Detrow sat down with Cal Newport, a computer science professor at Georgetown University and a contributing writer for The New Yorker. Newport, who has closely followed the evolution of large language models, offered a candid assessment: “It’s a great piece of technology, but it was not a transformative piece of technology, and that’s what we had been promised ever since GPT-4 came out, which is, the next major model was going to be the next major leap, and GPT-5 just wasn’t that.”
This sense of disappointment isn’t limited to ChatGPT alone. Newport explained that the entire field of large language models has hit a plateau. Initially, there was a clear trend: as models grew larger and were trained on more data, their performance improved dramatically. “There was an actual curve. It came out in a paper in 2020 that showed, this is how fast these models will get better as we make them larger, and GPT-3 and GPT-4 fell right on those curves. So we had a lot of confidence in the AI industry that, yeah, if we keep getting bigger, we’re going to keep moving up this very steep curve. But sometime after GPT-4, the progress fell off that curve and got a lot flatter,” Newport said, as reported by NPR.
The implications are significant. For years, the prevailing wisdom was that AI would continue to leap forward, eventually achieving superhuman intelligence through sheer scale. But Newport pointed out that this assumption is no longer tenable. “Essentially, the idea that simply making the model bigger and training it longer is going to make it much smarter—that has stopped working across the board. We first started noticing this around late 2023, early 2024. All of the major large language models right now have shifted to another way of getting better. They’re focusing on what I call post-training improvements, which are more focused and more incremental, and all major models from all major AI companies are focused on this more incremental approach to improvement right now.”
This shift in strategy—from pre-training (making models bigger and training them longer) to post-training (incremental, targeted enhancements)—reflects a broader reckoning within the industry. Newport used a simple metaphor to explain the change: “If pre-training gives you, like, a car, post-training soups up the car. And what has happened is we’ve turned our attention in the industry away from pre-training and towards post-training, so less trying to build a much better car and more focused on trying to get more performance out of the car we already have.”
But why are so many generative AI pilots failing in the workplace? The MIT report’s headline-grabbing 95% failure rate is not, in Newport’s view, particularly surprising. “What we were hoping was going to happen with AI in the workplace was the agentic revolution, which was this idea that maybe language models would get good enough that we could give them control of software, and then they could start doing lots of stuff for us in the business context. But the models aren’t good enough for that. They hallucinate. They’re not super reliable. They make mistakes or make odd behavior. And so these tools we’ve been building on top of language models—as soon as we leave very narrow applications where language models are very good, these more general business tools, they’re just not very reliable yet.”
This reality check is forcing a major rethink among both AI companies and the businesses eager to adopt their tools. The capital expenditure required to build and train massive AI models is, as Newport described, “astonishingly large.” To justify investments that can run into the hundreds of billions of dollars, companies need to find truly lucrative applications. Yet, so far, the killer app that can reliably deliver such returns remains elusive.
For workers and consumers, the slowdown in AI progress may come as a relief. Dire predictions about AI-driven mass unemployment—such as Dario Amodei’s warning that up to 20% of jobs could vanish, or that half of all new white-collar jobs could be automated in the near future—now appear far-fetched. “The technology is not there, and we do not have a route for it to get there in the near future. The farther future is a different question, but I do not think those scenarios of doom we’ve been hearing over the last six months or so—I think right now, they’re seeming unrealistic,” Newport told NPR.
So, what does the future hold? Rather than chasing ever-larger models, the industry is pivoting toward building bespoke tools tailored to specific use cases. Newport predicts, “Instead of just having this focus on making the models bigger and bigger and maybe you just access them through a chat interface, now we’re going to have to have a lot more attention on building bespoke tools on top of these foundation models for specific use cases. So I actually think the footprint in regular users’ lives is going to get more useful because you might get a tool that’s more custom fit for your particular job.”
Of course, challenges remain. Language models still generate misinformation, facilitate fraud, and produce content that can clutter the internet. But Newport sees a silver lining: “There’s still plenty of things to be worried about. Language models, as we have them today, can do all sorts of things that are a pain. It’s generating slop for the internet. It makes it much easier to have persuasive misinformation. The fraud possibilities are explosive. All of these things are negatives. But I’ll probably just get some better tools in the near future as just an average user. That’s not necessarily so bad.”
As the AI industry recalibrates its ambitions, the conversation has shifted from breathless forecasts of imminent superintelligence to a more nuanced, sober assessment of where the technology stands—and where it might actually deliver value. For now, the hype has given way to hard questions, and the answers will shape not only the future of AI, but the way we all work and live.
