Where RL Will Take Search — Maximilian-David Rumpf, SID.ai

This talk argues that reinforcement learning (RL) is the future of search optimization for AI agents. The speaker diagnoses the limitations of traditional search pipelines, which are characterized by fixed decisions and an inability to act on suboptimal results. RL is presented as a natural fit for search due to its verifiable reward signal and the ability to grind the environment at high speeds. The talk highlights that specialized RL models can achieve search speeds twenty times faster and at one hundredth the cost of frontier models, while also preventing bad results from entering the main context.