LLM judges have emerged as a key approach to evaluating open-ended scientific outputs that resist automated verification, yet their faithfulness in reference-free settings remains underexplored, particularly for fields such as novel materials synthesis where no ground-truth reference exists. Such judges produce false positives that accept scientifically invalid recipes and false negatives that penalize semantically equivalent ones in unfamiliar notation. We introduce ASSAY, a probe suite built on domain-grounded augmentations of ground-truth synthesis recipes, and reveal that these two failure modes form a systematic Stability–Validity trade-off across thirteen judge models. To resolve this trade-off, we train SynthJudge-8B via DPO on contrastive pairs derived from the same augmentations, improving judgment accuracy from 51.4% to 85.1% while maintaining agreement with domain experts. We release our probe suite and trained model to support faithful automated evaluation in scientific fields beyond materials synthesis at https://github.com/SoyeonHH/ASSAY.

Language models (LMs) are vulnerable to exploitation for adversarial misuse. Training LMs for safety alignment is extensive, making it hard to respond to fast-developing attacks immediately, such as jailbreaks. We propose self-refine with formatting that achieves outstanding safety even in non-safety-aligned LMsand evaluate our method alongside several defense baselines, demonstrating that it is the safest training-free method against jailbreak attacks.Additionally, we proposed a formatting method that improves the efficiency of the self-refine process while reducing attack success rates in fewer iterations. We observed that non-safety-aligned LMs outperform safety-aligned LMs in safety tasks by giving more helpful and safe responses.In conclusion, our findings can achieve less safety risk with fewer computational costs, allowing non-safety LM to be efficiently utilized in real-world service.

Text-to-SQL systems have become crucial for translating natural language into SQL queries in various industries, enabling non-technical users to perform complex data operations. The need for accurate evaluation methods has increased as these systems have grown more sophisticated. However, the Execution Accuracy (EX), the most prevalent evaluation metric, still shows many false positives and negatives. Thus, this paper introduces FLEX(False-Less EXecution), a novel approach to evaluating text-to-SQL systems using large language models (LLMs) to emulate human expert-level evaluation of SQL queries. Our metric improves agreement with human experts (from 62 to 87.04 in Cohen’s kappa) with comprehensive context and sophisticated criteria. Our extensive experiments yield several key insights: (1) Models’ performance increases by over 2.6 points on average, substantially affecting rankings on Spider and BIRD benchmarks; (2) The underestimation of models in EX primarily stems from annotation quality issues; and (3) Model performance on particularly challenging questions tends to be overestimated. This work contributes to a more accurate and nuanced evaluation of text-to-SQL systems, potentially reshaping our understanding of state-of-the-art performance in this field.

Caution: This paper includes offensive words that could potentially cause unpleasantness. The fast-paced evolution of generative language models such as GPT-4 has demonstrated outstanding results in various NLP generation tasks. However, due to the potential generation of offensive words related to race or gender, various Controllable Text Generation (CTG) methods have been proposed to mitigate the occurrence of harmful words. However, existing CTG methods not only reduce toxicity but also negatively impact several aspects of the language model’s generation performance, including topic consistency, grammar, and perplexity. This paper explores the limitations of previous methods and introduces a novel solution in the form of a simple Gated Toxicity Avoidance (GTA) that can be applied to any CTG method. We also evaluate the effectiveness of the proposed GTA by comparing it with state-of-the-art CTG methods across various datasets. Our findings reveal that gated toxicity avoidance efficiently achieves comparable levels of toxicity reduction to the original CTG methods while preserving the generation performance of the language model.