• Prefill/decode disaggregation improves latency-throughput tradeoff for large language model serving. • Energy consumption remains high; autoscaling is too coarse-grained for rapi
• FineRef introduces fine-grained error reflection for citation mismatch and irrelevance in long‑form LLM generation. • Two‑stage training: supervised fine‑tuning with attempt‑refl
• LLM benchmarks often misrepresent true model capabilities due to contamination and annotator errors. • Construct validity is essential to ensure benchmarks truly measure desired