Professor Questions Open Source Dominance in Non-Frontier AI

Michiel Bakker, an MIT assistant professor and Google DeepMind researcher, posted on X questioning assumptions that a large open source ecosystem will dominate non-frontier models while frontier labs focus on the most capable ones. He suggested that frontier labs' potential edges in automated researchers, compute, and teacher models for distillation could allow them to build cheaper models and achieve higher margins, potentially undercutting providers like Baseten and Together as a concern rather than a certainty. Herbie Bradley replied that open source inference providers avoid amortizing development costs and that frontier labs have not led in sub-frontier models so far. Bakker agreed the points are fair but wondered if the current division of labor is stable.