Get the latest tech news
Baba Is Eval
Making Language Models Play Baba is You - fi-le.net
The game is turn-based, meaning the number of turns required to solve a level naturally serves as a more fine-grained metric beyond accuracy. These write to one of four predefined storage files (such as "level"), in an INI format with sections delineated by[group] followed with key=value pairs on each new line. Manually solving the level we looked at above, we find a solution rrrrrrddddrrrrdldllluuuuurulllllllrrrrrrddddddrrruldlluuuuurullllllrrrrrrddddrdldluuuuurulllllullldlllllddlldluldddrruldlluurddddldrrrrrlluuurrrruurrrrllllllullld, where each letter stands for left, right, up or down, giving 324 bits.
Or read this on Hacker News