Generalizability of hash-conditioning beyond minimal algorithmic tasks

Ascertain whether hash-conditioning remains effective for real-world open-ended tasks beyond the four minimal algorithmic creativity benchmarks introduced in the paper, including tasks such as problem design, analogy generation, and idea discovery.

Background

Hash-conditioning shows notable benefits on the paper’s controlled algorithmic tasks, such as Sibling Discovery, Triangle Discovery, Circle Construction, and Line Construction.

The authors explicitly state uncertainty about whether these gains transfer to more complex, realistic tasks, calling for evaluation of external validity.