May 11, 2026 · 1 min read

Reward function

Good article here https://x.com/vansickn/status/2053644179352297841

A black background with white text discussing why code is often treated as a mere conduit to a completed task, attributed to extensive RL training. Below the text is a diagram labeled "FIGURE 2 WHAT THE REWARD FUNCTION SEES," showing a flow from "prompt" to "completed" with a wavy line, and a caption below reading "who cares how we got here!"

View original