Enterprise AI investments risk yielding homogeneous capabilities if they continue to rely on a shrinking number of foundation models. According to Dr Yichuan Zhang, CEO and co-founder of Boltzbit, writing in TechRadar Pro, most of today's AI products are at their core the same because they sit on top of a small number of foundation models trained on broadly similar datasets. The result, he writes, is a market of apparent variety masking underlying homogeneity with different interfaces but increasingly the same intelligence layer.
The Growing Homogenization Risk
The danger, according to Zhang, is that if the current trajectory holds, this small number of model providers will define not just the capabilities of AI systems but the boundaries of innovation and who participates in it. He states, "The wings of progress are being clipped." A world where every AI product behaves the same because it sits upon the same underlying intelligence is a world where differentiation erodes. Individuality, at both an organizational and personal level, becomes harder to express through technology.
Zhang points to nuclear fusion as an example of AI's potential: long thought to be always 30 years away, it is now being modeled and optimized with AI in ways that materially change the pace of experimentation. But the benefits of such breakthroughs risk being concentrated among a few providers if the AI ecosystem remains centralized.
The Case for Live Learning
To avoid this lost opportunity, Zhang argues for a shift at the most critical level: the intelligence layer. He writes that individuals and organizations need the ability to shape, adapt, and own the models that power their applications. This is where live learning becomes critical. Static models, no matter how large, are nothing but snapshots that improve through periodic retraining cycles controlled by their providers. Live learning models, by contrast, evolve continuously in production, incorporating new data and allowing users to control what and how the models learn.
Zhang frames this as the difference between renting intelligence and owning it. He notes that technology using live learning models already exists in production, but the real task is making it accessible at scale in a usable and sustainable way.
| Aspect | Static Foundation Models | Live Learning Models |
|---|---|---|
| Update cycle | Periodic retraining by provider | Continuous evolution in production |
| User control | Limited to provider's roadmap | Users shape what and how models learn |
| Result | Rented intelligence | Owned intelligence |
How Intelligence Evolves in Practice
A live environment, Zhang explains, is crucial to understanding how users interact with, adapt to, and derive value from live learning systems. It provides the ability to observe how intelligence evolves when placed directly in the hands of users. What matters is not the first iteration but what it reveals: how users shape their models, what kinds of feedback loops emerge, and how intelligence behaves when it is no longer centrally controlled. Those insights will inform what comes next.
Implications for the Enterprise
For organizations that are becoming AI-native—embedding AI into their core operating models rather than adding it as a feature—the current trajectory matters enormously. If the ecosystem continues to be dominated by a small number of foundation model providers, Zhang warns that businesses face a future of powerful but standardized AI-assisted experiences. The alternative is the democratization of live learning: a world where any individual or organization can train and own the intelligence layer powering their applications, and where that intelligence continues to evolve through feedback and additional data.
In that world, the performance, management, and evolution of AI systems are no longer tied to the roadmap of a handful of providers. Zhang concludes, "That is the difference between participating in the AI-native era and inheriting someone else's version of it."
Sources:
