Entity-fact unlearning in large language models (LLMs) aims to remove the binding between a target entity and a forgotten fact, producing a model that behaves like one retrained without the forgotten data while preserving neighboring knowledge and overall model utility. We frame this problem as distribution shaping: such a retrained model would yield a plausible distribution at forget queries, fluent, semantically appropriate, and reflective of residual knowledge, which we take as the appropriate target. Existing objectives provide only narrow token-level or semantically uninformative supervision, falling short of this target. We propose Counterfactual Self-Distillation (CSD), a self-distillation framework that approximates the retrained distribution by constructing a counterfactual teacher through type-consistent entity substitution. The resulting teacher is decoupled from the target identity while preserving type-level relational structure, enabling directed probability reallocation without an external oracle. Experiments on TOFU and RWKU show that CSD achieves a strong forget-utility trade-off, including the best overall forgetting on RWKU and strong performance on TOFU.

Although Large Language Models (LLMs) have made remarkable progress, current preference optimization methods still struggle to align directional consistency while preserving reasoning diversity. To address this limitation, we propose Directional-Groupwise Preference Optimization (DGPO), a lightweight framework that aggregates supervision signals at the group level and explicitly models direction-aware alignment through multi-candidate comparisons. DGPO organizes forward and reverse question-answer instances into structured sets and optimizes a margin-based likelihood objective that separates coherent reasoning paths from inconsistent alternatives. This groupwise formulation captures richer relative information than pairwise objectives and reinforces consistency across diverse reasoning pathways. Empirical results show that our constructed reverse data yields a 3.2% average improvement across five benchmarks, while DGPO further delivers consistent gains across multiple datasets and model families, achieving average accuracy improvements of up to 3.6%. Our code and data are available at https://github.com/Demi-deng2/DGPO.

Language agents increasingly rely on reusable skills, but post-failure repair is often handled by opaque one-shot reflection: a model generates a skill patch without explicitly maintaining how failure explanations relate to candidate repairs or how unsuccessful retests should influence later edits. We introduce RESKILL, a structured repair framework that maintains an explicit repair state across repair rounds. Given a failed rollout, the framework links failure hypotheses to candidate skill patches, selects local repairs through coverage-based attribution, retests the edited skill set in the environment, and uses retest outcomes to guide subsequent repair updates. The language model supplies structured repair factors, while the repair procedure records them, compares local skill patches by how well they address active failure explanations, and carries unsuccessful retest outcomes into later repair rounds. We evaluate RESKILL on ALFWorld and TextCraft across three model sizes under fixed repair budgets. RESKILL obtains the strongest final success in all six benchmark–model settings, improving average final success by 3.7 percentage points over direct repair and 3.3 points over hypothesis-conditioned repair. These results suggest that explicit attribution alone is insufficient; durable improvement emerges when attribution is integrated with repair selection and persistent retest-conditioned update.

Existing work has shown that o1-level performance can be achieved with limited data distillation, but most existing methods focus on unidirectional supervised fine-tuning (SFT), overlooking the intricate interplay between diverse reasoning patterns. In this paper, we construct r1k, a high-quality reverse reasoning dataset derived by inverting 1,000 forward examples from s1k, and examine how SFT and Direct Preference Optimization (DPO) affect alignment under bidirectional reasoning objectives. SFT on r1k yields a 1.6%–6.8% accuracy improvement over s1k across evaluated benchmarks. However, naively mixing forward and reverse data during SFT weakens the directional distinction. Although DPO can partially recover this distinction, it also suppresses less preferred reasoning paths by shifting the probability mass toward irrelevant outputs. These findings suggest that mixed reasoning data introduce conflicting supervision signals, underscoring the need for robust and direction-aware alignment strategies. Our code and data are available at: https://github.com/16demi/ReasonAlign-analysis.