Spreadsheet-RL: Advancing Large Language Model Agents on Realistic Spreadsheet Tasks via Reinforcement Learning

Paper 1 presents a more scientifically novel contribution by introducing neuroanatomical inductive priors for brain decoding, addressing a fundamental performance-fidelity trade-off in fMRI research. Its innovations (SRST, SG-MoE) are methodologically rigorous with causal ablations, achieve state-of-the-art results with dramatic efficiency gains (10 vs 600 epochs), and have broad implications for neuroscience and clinical applications. Paper 2, while practically useful, is more incremental—applying RL fine-tuning to spreadsheet automation with moderate performance gains on a narrow application domain, representing engineering advancement rather than fundamental scientific insight.

Paper 2 addresses a highly ubiquitous real-world problem (spreadsheet automation) with a novel RL fine-tuning framework and environment for LLM agents. Its significant performance improvements and broad applicability across diverse industries give it much higher potential for widespread adoption and cross-disciplinary impact compared to Paper 1's niche, negative-result benchmark in multi-agent reinforcement learning.

Paper 1 has higher impact potential due to a more actionable, end-to-end contribution: an RL fine-tuning framework, a realistic Excel-based environment (Spreadsheet Gym), and new curated benchmarks enabling reproducible progress on a high-value workplace domain. It demonstrates sizable empirical gains and directly targets automation of ubiquitous spreadsheet workflows, with clear real-world and cross-domain applications (finance, ops, supply chain). Paper 2 provides valuable diagnostic evaluation of VLM numeric grounding, but is primarily an analysis/benchmarking effort with limited immediate performance improvements, making near-term applied impact narrower.

Paper 2 likely has higher scientific impact because it tackles a timely, broadly relevant security problem for memory-augmented LLM agents (post-hoc auditing after harm), with clear cross-domain implications for agent safety, governance, and deployment. Its causal attribution + structural anomaly approach is conceptually novel and generalizable across agent architectures and applications, and it reports large risk reductions under a realistic attack model. Paper 1 is valuable and application-focused, but its impact is narrower (spreadsheet automation) and more incremental (RL fine-tuning + benchmark/environment engineering) with modest performance gains.

While Paper 2 offers significant practical utility through spreadsheet automation, Paper 1 addresses fundamental challenges in AI reasoning and formal mathematics. By exploring iteratively self-improving models and neurosymbolic scaffolding for formal proof optimization, Paper 1 contributes to the critical and rapidly evolving frontier of AI capabilities in logic, reasoning, and synthetic data generation, which generally yields broader scientific and theoretical impact across artificial intelligence.

Spreadsheet-RL introduces a novel RL fine-tuning framework with concrete, measurable improvements on practical tasks, a new benchmark dataset, and a training environment (Spreadsheet Gym) that enables reproducible research. Its real-world applicability to ubiquitous spreadsheet workflows gives it broad impact potential. While AttuneBench addresses an important gap in emotional intelligence evaluation with a thoughtful methodology, it is primarily a benchmark/evaluation contribution without a training methodology. Spreadsheet-RL's combination of framework, benchmark, and demonstrated performance gains across general and domain-specific tasks suggests wider methodological influence and practical adoption.

Paper 2 likely has higher impact: it targets a ubiquitous real-world platform (Excel) with broad cross-industry utility, introduces an RL fine-tuning framework plus a realistic environment (Spreadsheet Gym) and a new benchmark/dataset, enabling reproducible progress and follow-on work. The methodological contribution (RL in a tool-rich, multi-turn setting) is timely for LLM agents and computer-use research, with potential to influence agent evaluation and training beyond spreadsheets. Paper 1 is useful but more niche (story illustration) and largely prompt-layer orchestration without new training or benchmarks at comparable breadth.

Spreadsheet-RL addresses a highly practical and widespread problem (spreadsheet automation) with a novel RL-based framework, including a new benchmark, training environment, and data pipeline. Its real-world applicability to billions of spreadsheet users gives it enormous potential impact. While ST-SimDiff offers a clever training-free video token reduction method, it represents more of an incremental improvement in the already crowded video understanding efficiency space. Spreadsheet-RL opens a relatively underexplored research direction combining RL with domain-specific tool use, with broader implications for LLM-based data interface agents.

Paper 1 is likely to have higher scientific impact due to its methodological contribution (an RL fine-tuning framework plus a realistic Excel “gym”), creation of scalable data/benchmark resources, and clear, measurable performance gains on practical tasks. Its real-world applicability (spreadsheet automation in finance, supply chain, etc.) is immediate and broadly relevant to enterprise workflows and agentic AI. Paper 2 is timely for AI safety and cognitive alignment, but is primarily an evaluation/probing study with narrower direct application and less of a reusable systems/dataset/tooling advance.

ResDreamer introduces a novel hierarchical world model architecture with residual reconstruction for self-supervised visual reasoning in 3D environments—a fundamental contribution to reinforcement learning and world models. Its principled, domain-agnostic design following the 'Bitter Lesson' and demonstrated scalability make it broadly impactful across RL, robotics, and embodied AI. Paper 2, while practically useful for spreadsheet automation, is more application-specific, incremental in its use of RL fine-tuning for LLM agents, and addresses a narrower problem domain with limited broader scientific novelty.