@misc{indiciae2914e91192a1,
  title = {LLMs can construct powerful representations and streamline sample-efficient supervised learning},
  author = {Ilker Demirel and Lawrence Shi and Zeshan Hussain and David Sontag},
  year = {2026},
  url = {https://arxiv.org/abs/2603.11679},
  note = {Source identifier: 2603.11679}
}