Foundation models are the emperor’s new clothes of AI, and it’s time we called them out. Sure, they can generate coherent sentences and even pass the Turing test in short bursts, but are they truly the revolutionary leap forward they’re hyped up to be? Not really. These models are essentially supercharged parrots, regurgitating patterns they’ve seen in their training data. They lack genuine understanding or creativity, and their “intelligence” is a facade built on statistical correlations.

I’ve seen this firsthand. A friend of mine, a software engineer, was tasked with integrating a foundation model into their product. The model could spit out code snippets that looked impressive but often missed the mark on actual functionality. It was a band-aid solution that ended up creating more work in the long run. And let’s not even get started on the ethical dilemmas—these models often perpetuate biases present in their training data, leading to discriminatory outcomes that are hard to trace and even harder to fix.

The hype around foundation models is driven by the same forces that have inflated every tech bubble: the allure of quick fixes and the fear of missing out. But just like those bubbles, this one is bound to burst. The real transformative power of AI will come from models that can truly understand and reason, not just mimic.

Foundation models are not the future; they’re a stepping stone, and it’s time we started treating them as such.