In this monologue, Toju Duke outlines five major trends he expects to shape artificial intelligence by 2026. First, the AI arms race will intensify as generative models keep gaining capabilities, with rapid leaps from multiple competing players. Second, a wave of AI sovereignty initiatives will emerge, as countries such as Japan, Saudi Arabia, EU members, Russia, India, and South Korea develop locally‑trained large language models in their native languages. Third, the community will achieve a deeper grasp of LLM behavior, especially the phenomenon of semantic leakage—where fine‑tuning on new data introduces inaccuracies and confused model states. Fourth, efforts to curb bias and representational harms will accelerate, exemplified by Duke’s own DiverseAI project that will release a novel 5‑million‑image dataset designed for factual and diverse representation. Finally, the push toward artificial general intelligence will gain momentum through advances in world models, which simulate real‑world causal dynamics inside AI agents, enabling more reliable prediction and planning.
