@misc{indiciae2fe232e8a6af,
  title = {Autoregressive Language Models are Secretly Energy-Based Models: Insights into the Lookahead Capabilities of Next-Token Prediction},
  author = {Mathieu Blondel and Michael E. Sander and Germain Vivier-Ardisson and Tianlin Liu and Vincent Roulet},
  year = {2026},
  url = {https://arxiv.org/abs/2512.15605},
  note = {Source identifier: 2512.15605}
}