Can David Beat Goliath? On Multi-Hop Reasoning with Resource-Constrained Agents

While reinforcement learning (RL) has empowered multi-turn reasoning agents with retrieval and tools, existing successes largely depend on extensive on-policy rollouts in high-cost, high-accuracy regimes. Under realistic resource constraints that cannot support large models or dense explorations, however, small language model agents fall into a low-cost, low-accuracy regime, where limited rollout budgets lead to sparse exploration, sparse credit assignment, and unstable training. In this work, we challenge this trade-off and show that small language models can achieve strong multi-hop reasoning under resource constraints. We introduce DAVID-GRPO, a budget-efficient RL framework that (i) stabilizes early learning with minimal supervision, (ii) assigns retrieval credit based on evidence recall, and (iii) improves exploration by resampling truncated near-miss trajectories. Evaluated on agents up to 1.5B parameters trained on only four RTX 3090 GPUs, DAVID-GRPO consistently outperforms prior RL methods designed for large-scale settings on six multi-hop QA benchmarks. These results show that with the right inductive biases, small agents can achieve low training cost with high accuracy.

Analyzing the Effectiveness of Listwise Reranking with Positional Invariance on Temporal Generalizability

Benchmarking the performance of information retrieval (IR) methods are mostly conducted within a fixed set of documents (static corpora). However, in real-world web search engine environments, the document set is continuously updated and expanded. Addressing these discrepancies and measuring the temporal persistence of IR systems is crucial. By investigating the LongEval benchmark, specifically designed for such dynamic environments, our findings demonstrate the effectiveness of a listwise reranking approach, which proficiently handles inaccuracies induced by temporal distribution shifts. Among listwise rerankers, our findings show that ListT5, which effectively mitigates the positional bias problem by adopting the Fusion-in-Decoder architecture, is especially effective, and more so, as temporal drift increases, on the test-long subset.