Pushing the Boundaries of Multiple Choice Evaluation to One Hundred Options

Multiple choice evaluation is widely used for benchmarking large language models, yet near ceiling accuracy in low option settings can be sustained by shortcut strategies that obscure true competence. Therefore, we propose a massive option evaluation protocol that scales the candidate set to one hundred options and sharply reduces the impact of chance performance. We apply this framework to a Korean orthography error detection task where models must pick the single incorrect sentence from a large candidate set. With fixed targets and repeated resampling and shuffling, we obtain stable estimates while separating content driven failures from positional artifacts. Across experiments, results indicate that strong performance in low option settings can overstate model competence. This apparent advantage often weakens under dense interference at high $N$, revealing gaps that conventional benchmarks tend to obscure. We identify two failure modes, semantic confusion and position bias toward early options under uncertainty. To isolate the effect of context length, we run padding controlled and length matched tests, which suggest that the main bottleneck is candidate ranking rather than context length. Together, these findings support massive option evaluation as a general framework for stress testing model reliability under extreme distractor density, beyond what low option benchmarks can reveal.

BEAT2AASIST model with layer fusion for ESDD 2026 Challenge

Recent advances in audio generation have increased the risk of realistic environmental sound manipulation, motivating the ESDD 2026 Challenge as the first large-scale benchmark for Environmental Sound Deepfake Detection (ESDD). We propose BEAT2AASIST which extends BEATs-AASIST by splitting BEATs-derived representations along frequency or channel dimension and processing them with dual AASIST branches. To enrich feature representations, we incorporate top-k transformer layer fusion using concatenation, CNN-gated, and SE-gated strategies. In addition, vocoder-based data augmentation is applied to improve robustness against unseen spoofing methods. Experimental results on the official test sets demonstrate that the proposed approach achieves competitive performance across the challenge tracks.

KAIO: A Collection of More Challenging Korean Questions

With the advancement of mid/post-training techniques, LLMs are pushing their boundaries at an accelerated pace. Legacy benchmarks saturate quickly (e.g., broad suites like MMLU over the years, newer ones like GPQA-D even faster), which makes frontier progress hard to track. The problem is especially acute in Korean: widely used benchmarks are fewer, often translated or narrow in scope, and updated more slowly, so saturation and contamination arrive sooner. Accordingly, at this moment, there is no Korean benchmark capable of evaluating and ranking frontier models. To bridge this gap, we introduce KAIO, a Korean, math-centric benchmark that stresses long-chain reasoning. Unlike recent Korean suites that are at or near saturation, KAIO remains far from saturated: the best-performing model, GPT-5, attains 62.8, followed by Gemini-2.5-Pro (52.3). Open models such as Qwen3-235B and DeepSeek-R1 cluster falls below 30, demonstrating substantial headroom, enabling robust tracking of frontier progress in Korean. To reduce contamination, KAIO will remain private and be served via a held-out evaluator until the best publicly known model reaches at least 80% accuracy, after which we will release the set and iterate to a harder version.

Controlling Language Confusion in Multilingual LLMs

Large language models often suffer from language confusion, a phenomenon where responses are partially or entirely generated in unintended languages. This can critically impact user experience in low-resource settings. We hypothesize that conventional supervised fine-tuning exacerbates this issue because the softmax objective focuses probability mass only on the single correct token but does not explicitly penalize cross-lingual mixing. Interestingly, by observing loss trajectories during the pretraining phase, we observe that models fail to learn to distinguish between monolingual and language-confused text. Additionally, we find that ORPO, which adds penalties for unwanted output styles to standard SFT, effectively suppresses language-confused generations even at high decoding temperatures without degrading overall model performance. Our findings suggest that incorporating appropriate penalty terms can mitigate language confusion in low-resource settings with limited data.