December 26th.Anthropic Co-FounderJack Clark stated.Thanks to the scaling of traditional models and the proliferation of new models and approaches, the AI field continues to roar down the highway in 2025.
In his Import AI blog, Clarke countered the notion that AI development has reached its limits by stating that the o3 model is not simply increasing the size of the model, but rather utilizing reinforcement learning and additional computational power to open up a whole new set of possibilities for scaling.
Clark predicts that major companies will further drive the rapid growth of the AI field in 2025 through a combination of extending the underlying models and integrating new approaches in the training and inference process.
Clark believes that most people are not prepared for the rapid developments that are coming, warning, "I don't think anyone at this stage can predict the rapid growth of AI in the future."
Citing the blog post, 1AI also highlighted computational cost as a major challenge, with the high-computing version of the OpenAI o3 model requiring 170 times the computational power of its base version, and the low-computing version of o3 costing several times as much as the o1 model, so greater arithmetic requirements and high costs may further drive AI hardware development.