DeepMind paper proves sequence-level supervised losses for autoregressive models can be fundamentally flawed

On November 17, Google DeepMind researcher Mathieu Blondel and colleagues published a paper on arXiv, "On the Difficulty of Sequence-Level Supervised Learning for Autoregressive Models," systematically demonstrating that sequence-level supervised training of autoregressive models faces fundamental difficulties.

Confirmed

Why it matters

All of the above comes from Blondel's own paper-explainer thread on X (m1–m3 are parts of the same thread; m4 and m5 are paraphrases).

2026-11-17 ~ 2026-11-17 · 5 related posts

Primary sources

1 near-duplicate retellings: mblondel_ml