J. Zhang, “Do Language Models Prefer Better Demonstrations or Easier Output Tokens?”, CAI, vol. 3, no. 5, pp. 1–10, Oct. 2026, doi: 10.70267/cai.v3n5.0110.