4 papers
Do Coding Agents Deceive Us? Detecting and Preventing Cheating via Capped Evaluation with Randomized Tests
Thanawat Lodkaew, Johannes Ackermann, Soichiro Nishimori +3
A growing failure mode in agent evaluation and training is that models can achieve high evaluation scores by exploiting shortcuts instead of solving the intended task, producing de…
On Symmetric Losses for Robust Policy Optimization with Noisy Preferences
Soichiro Nishimori, Yu-Jie Zhang, Thanawat Lodkaew +1
Optimizing policies based on human preferences is key to aligning language models with human intent. This work focuses on reward modeling, a core component in reinforcement learnin…
CapBencher: Give Your LLM Benchmark a Built-in Alarm for Test-Set Overfitting
Takashi Ishida, Thanawat Lodkaew, Ikko Yamane
Publishing a large language model (LLM) benchmark (especially its ground-truth answers) on the Internet risks contaminating future LLMs and enabling evaluation gaming: it may be un…
THaLLE: Text Hyperlocally Augmented Large Language Extension -- Technical Report
KBTG Labs, Danupat Khamnuansin, Atthakorn Petchsod +6
Recent advancements in Large Language Models (LLMs) have revealed new capabilities and opportunities across the technological landscape. However, the practicality of very large LLM…