For three years, raw compute defined the capability frontier. But internal evals from multiple top labs now show that scaling Transformers alone is approaching diminishing returns on core reasoning and planning tasks. We map six parallel research paths — and speak with leading researchers — to understand what comes next.
The AGI Route Wars: Scaling Laws Are Stalling, and World Models Are Taking Over
For three years, raw compute defined the capability frontier. But internal evals from multiple top labs now show that scaling Transformers alone is approaching diminishing returns on core reasoning and planning tasks. We map six parallel research paths — and speak with leading researchers — to understand what comes next.
