LLM 集成增益从哪来：候选价值、多样性与聚合的拆解
Where Do LLM Ensemble Gains Come From? Decomposing Candidate Value, Diversity, and Aggregation
arXiv 论文用 MultiHopRAG、FRAMES 证明：评 LLM 集成收益不先定对照，数字能差 8 个百分点。
论文将 LLM 聚合器放在 Direct、Single-slot、Duplicate、Multi-candidate 四种匹配对照条件下比较。在 MultiHopRAG（k=2）上，候选总增益 CIV 可代数拆分为额外候选对比 DSV 与候选上下文对比 CCV，但拆分结果依赖所选对照。MultiHopRAG 上以 Duplicate 与 Single-slot 为参照测得的增益差 Rep=-8.33 个百分点（区间 [-13.25, -3.75]）。在 n=200、k=3、输出上限 4096 的 FRAMES 强化面板扩展中，配对 Delta Rep=+9.67 pp [+6.00, +13.50]，且 Rep 在原面板为 -6.33、强化面板转为 +3.33。作者据此提出：把增益归因于额外候选前，集成评估必须先声明参照对照。
Where Do LLM Ensemble Gains Come From? Decomposing Candidate Value, Diversity, and Aggregation
LLM ensemble gains are often measured against a single baseline, but the value attributed to additional candidates can change with the comparison used. We study an LLM aggregator under four matched conditions: Direct (no candidates), Single-slot (one candidate in a multi-slot interface), Duplicate (one candidate repeated), and Multi-candidate (separately generated candidates). Under either non-diverse control, overall candidate gain (CIV) splits algebraically into an additional-candidate contrast (DSV) and a candidate-context contrast (CCV); the split, but not CIV, depends on the chosen control. On MultiHopRAG model diversity (k=2), the repetition gap Rep=-8.33 pp [-13.25, -3.75] quantifies the difference between gains measured against Duplicate and Single-slot. On a matched FRAMES stronger-panel extension (n=200, k=3, aggregator output limit 4096), paired Delta Rep=+9.67 pp [+6.00, +13.50]; within panels, Rep changes sign (original -6.33; stronger +3.33). In FRAMES evidence aggregation, candidate-context contrasts remain strongly negative under both non-diverse controls. Our results show why ensemble evaluations should state the reference control before attributing gains to additional candidates.
