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

• 类型：arxiv
• 标识：2608.17744
• 链接：https://arxiv.org/abs/2608.17744
• 主分类：evaluation
• 形态：benchmark
• 被引：0
• 被引来源：Semantic Scholar
• S2被引：0
• 影响力被引：0
• TLDR：Take three frontier mixture-of-experts models and fine-tune them to reason in a low-resource language and propose six behavioural dimensions that make changes measurable, each gated to reject any metric that correlates with output length, and report how their own instruments lied.
• 待LLM分类：否
• 标题中文：低资源语言下的思考：SFT 构建什么，RL 修复什么，准确率看不到什么
• TLDR中文：选取三个前沿混合专家模型在低资源语言上进行推理微调，提出六个可度量的行为维度，且每维度均设门拒绝任何与输出长度相关的指标，并报告其自家评测工具为何失效。
• 来源文件：
• /inbox/tom/_candidates/2026-08-21-agent-rag-longcontext-substack-candidates.json
• /inbox/tom/_candidates/2026-08-21-agent-rag-longcontext-candidates.json
• [S2 enrich]