Research map / Large language models / Language, safety and interpretability
SpreadsheetArena: Decomposing Preference in LLM Generation of Spreadsheet Workbooks
Srivatsa R Kundurthy, Clara Na, Michael Handley, Zach Kirshner, Chen Bo Calvin Zhang, Manasi Sharma, Emma Strubell, John Ling
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| control for 29 spreadsheet features in the Bradley-Terry model (ablation) | Claude Opus 4.5 rating · SpreadsheetArena | Elo: 1550 -> 1333 | Hurt |
| control for spreadsheet features (ablation) | Qwen3 30B rating · SpreadsheetArena | Elo: 692 -> 849 | Improved |
| control for spreadsheet features (ablation) | Gemini 3 Pro rating · SpreadsheetArena | Elo: 1325 -> 1268 | Hurt |
| higher text density in a spreadsheet | win probability · SpreadsheetArena | BT coefficient: +1.562 | Improved |
| higher formula error rate | win probability · SpreadsheetArena | BT coefficient: -1.338 | Hurt |
| add lookup functions, conditionals or embedded constants | win probability · SpreadsheetArena | significance: p = 0.73 / 0.28 / 0.55 | No clear effect |
| add cell borders | win probability · Creative & Generative prompts | BT coefficient: -0.87 | Hurt |
| add cell borders | win probability · Finance prompts | BT coefficient: +0.57 | Improved |
| follow finance color conventions | win probability · Finance prompts | BT coefficient: +1.63 | Improved |
| stratify BT model by prompt category instead of pooled (ablation) | feature-augmented BT model · SpreadsheetArena | likelihood ratio test: p < 10^-68 | Improved |
| higher failure rate on Prompt Miss tag | model win rate · SpreadsheetArena | Pearson r: r = -0.807 | Hurt |
| higher Numerical computation failure rate | model win rate · SpreadsheetArena | Pearson r: r = -0.002 | No clear effect |
Closest papers by meaning
- SpreadsheetBench: Towards Challenging Real World Spreadsheet Manipulation (NeurIPS 2024 · similarity 0.90)
- EnterpriseBench: Evaluating LLM Agents on End-to-End Spreadsheet Tasks in Finance (NeurIPS 2026 · similarity 0.88)
- FinBen: A Holistic Financial Benchmark for Large Language Models (NeurIPS 2024 · similarity 0.86)
- FLAME: A Small Language Model for Spreadsheet Formulas (AAAI 2024 · similarity 0.85)
- Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting (ICLR 2024 · similarity 0.85)
- MINT: Evaluating LLMs in Multi-turn Interaction with Tools and Language Feedback (ICLR 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
