Call for Papers
Workshop on Foundations of LLM Post-Training in Changing Environments @ NeurIPS 2026
Motivation and Timeliness
LLMs are inherently multitask systems, capable of performing a wide range of tasks using a single set of parameters. These models are adapted to specific downstream tasks and usage contexts through post-training, for example via fine-tuning on task-specific data, preference-based updates, or parameter-efficient adaptation methods.
In practice, downstream tasks addressed through post-training often do not remain fixed after deployment. Their definitions, data, and evaluation criteria can evolve over time as models are used in changing environments and usage contexts. This non-stationarity leads to repeated post-deployment modifications, such as continual or incremental updates. These practices are largely guided by heuristics and empirical validation, with limited theoretical understanding of when adaptation succeeds, when it fails, or how it affects previously learned capabilities.
This gap is particularly consequential in safety-critical settings. The International AI Safety Report, published in 2025 and 2026 and endorsed by several leading experts, identifies the lack of theoretical understanding as a key limitation of current AI safety approaches. As a result, empirical improvements alone are insufficient to reason about robustness under task evolution, unintended regressions, and the limits of safe adaptation. These challenges motivate the need for theoretical and statistical foundations that can support principled post-training in dynamic, real-world environments.
Workshop Goals and Topics
We bring together researchers to advance theoretical understanding of LLM post-training under evolving tasks. We welcome submissions on any topic related to post-training of LLMs in non-stationary or evolving environments, including but not limited to:
Preference feedback as data
Statistical modeling of preference signals, including identifiability questions, heterogeneous annotators, varying feedback strength, and principled treatments of noise and misspecification.
Robustness and valid inference
Conditions under which post-training updates are robust to modeling choices, data collection effects, or feedback errors, as well as methods for uncertainty quantification, calibration, and principled evaluation.
Adaptive data collection and feedback loops
Effects of sequential or adaptive data collection on post-training, including selection bias, feedback loops, active query design, and evolving standards or evaluation criteria.
Adaptation mechanisms and limits
Theoretical understanding of post-training mechanisms such as parameter-efficient adaptation, modular updates, or selective fine-tuning, including limits of adaptation, trade-offs with capability preservation, and safety-relevant failure modes.
Key Dates
| Milestone | Date | Notes |
|---|---|---|
| Submission Portal Opens | TBA | OpenReview |
| Abstract Registration Deadline | TBA | AoE |
| Paper Submission Deadline | September 2026 (TBA) | AoE — firm |
| Review Period | September – October 2026 | Double-blind |
| Author Notification | October 2026 (TBA) | |
| Camera-Ready Deadline | November 2026 (TBA) | Non-archival |
| Workshop Date | December 2026 | Paris, France |
Submission Instructions
Format
Papers should be submitted in PDF format using the NeurIPS 2026 style file. Long papers may be up to 8 pages excluding references and appendices. Short papers (position papers, work in progress) are up to 4 pages. There is no page limit on references or supplementary material.
Review Process
All submissions will undergo double-blind peer review. Authors should anonymise their submissions and avoid self-identifying references in the main paper. Reviewers will evaluate novelty, significance, technical quality, and clarity.
Dual-submission Policy
FLLMPT 2026 is a non-archival workshop. We welcome submissions of work currently under review at other venues and work that has previously been published, provided it has been updated or recontextualised for the workshop audience. Please disclose prior publication in the submission form.
Submission Portal
Submissions will be handled via OpenReview. The link will be published on this page when the portal opens.