Dongxi NLP argues a cheap System-1 should gate costly System-2 compute toward AGI

Dongxi NLP (@dongxi_nlp) argues that progress toward AGI depends less on adding compute than on a cheap System-1 that decides what an expensive System-2 should compute.

The post says every rollout, reasoning step and tool call is paid for before its usefulness is known, and that most do not help. The author links to an article making the same argument. The post presents this as a claim and gives no experiments or measured results.