Europe 2026
Scaling the Next Paradigm of Heterogeneous Intelligence — Adrian Bertagnoli, Callosum
Overview
This talk proposes a shift from homogeneous AI systems, which rely on scaling single models across identical hardware, to a new paradigm of heterogeneous intelligence. This approach leverages the inherent complexity of real-world problems by decomposing them into sub-problems that can be addressed by diverse models, workflows, and hardware working in concert. The core thesis is that this heterogeneity, when properly orchestrated, leads to more efficient, faster, and cheaper AI systems.
Who should watch
- AI Engineers and Builders exploring advanced system architectures.
- Product Managers and Designers seeking to understand future AI capabilities and efficiencies.
- Researchers interested in the co-evolution of AI models, software workflows, and hardware.
- Those facing challenges with scaling single, monolithic AI models for complex, multi-step tasks.
Key takeaways
- Homogeneous intelligence, based on scaling single models with more data and parameters, is becoming less relevant for inference compared to training.
- Mild heterogeneity is already emerging through mixture-of-experts architectures, multi-agent systems, and disaggregated hardware.
- The next paradigm involves a co-evolution of systems, hardware, and software, leading to deeply integrated heterogeneous intelligence.
- Heterogeneity is mathematically proven to outperform homogeneous systems for complex, multi-step problems by matching diverse sub-problems to specialized intelligences.
- Heterogeneous recursion extends concepts like recursive language models by mapping sub-contexts to different models and hardware, drastically reducing cost and increasing speed.
- In visual web navigation, a heterogeneous approach using a mix of open and closed video action language models outperformed state-of-the-art homogeneous models.
- Offloading simpler subtasks, like zooming, to less intelligent models significantly improves efficiency and reduces costs compared to using large, general-purpose models.
- The future of compute is seen as heterogeneous, mapping multi-agentic workloads onto diverse chips, moving beyond the era of massively parallel homogeneous compute.
Notable quotes
*Real-world problems are complex, multi-step, and open-ended. They decompose into sub-problems, which require vastly different types of intelligences.*
*The era of homogeneous scaled delivered extraordinary progress. What comes next is heterogeneous intelligence where models, workflows, and silicon co-evolve.*
*This is the worst our infrastructure will ever be.*
Unofficial community note. Prefer the recording for nuance.