Fine-tuning makes AI reason in Greek, improving transparency

Original: Thinking in a Low-Resource Language: What SFT Builds, What RL Fixes, What Accuracy Cannot See

Source: arxiv.org ↗