@article{featuresasrewardsscalablesupervisionforo,
  title = {Features as Rewards: Scalable Supervision for Open-Ended Tasks via Interpretability},
  author = {Aaditya Vikram Prasad and Connor Watts and Jack Merullo and Dhruvil Gala and Owen Lewis and Thomas McGrath and Ekdeep Singh Lubana},
  year = {2026},
  eprint = {2602.10067},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2602.10067},
}