NeurIPS 2026 workshop targets the pre-training to post-training gap
srush_nlp · x · 2026-08-04
NeurIPS 2026 workshop will study the transition from pre-training to post-training
A new NeurIPS 2026 workshop, “Transitioning from Pre-Training to Post-Training,” will take place on December 11, 2026 in Sydney as a full-day, in-person, non-archival event on OpenReview.
The workshop focuses on a gap the organizers say modern foundation-model research still lacks: a principled understanding of how pre-training and post-training relate. They want papers on:
- what a base model must already have for post-training to succeed
- how supervised fine-tuning, preference optimization, RL, self-improvement, and distillation reshape the model
- when and why post-training fails
- whether pre-training-side signals can predict post-training outcomes
- how data and optimization choices govern the transition between stages
The call for papers asks researchers to bring theory, empirical evidence, and benchmarks to turn what the organizers call “folklore” into science. The submission deadline is August 29, 2026.
More from Research
- Epoch AI’s MirrorCode benchmark sees Claude Fable solve C preprocessor and Pkl tasks — Jsevillamol · 2026-08-04
- A NeurIPS joke says authors waiting for rebuttals are basically Waiting for Godot — MelMitchell1 · 2026-08-04
- Researcher says RLVR gains on noisy labels are spurious, not real model improvement — StellaLisy · 2026-08-04
- Courant’s 1928 PDE methods foreshadowed electronic numerical computing — ylecun · 2026-08-04
- A trolley-problem video prefix is used to probe bias in model continuations — Kangwook_Lee · 2026-08-04
- First InterpScience workshop on LLM interpretability opens for NeurIPS 2026 — sivareddyg · 2026-08-04