PLUMB datasets: tasks, runs and traces Collection Locked 96-task benchmark, a corrected version, a leak-free split, audited SFT corpora, and the Dax-1 probes in its native patch contract. • 2 items • Updated about 11 hours ago
PLUMB datasets: tasks, runs and traces Collection Locked 96-task benchmark, a corrected version, a leak-free split, audited SFT corpora, and the Dax-1 probes in its native patch contract. • 2 items • Updated about 11 hours ago
PLUMB: grounding for spreadsheet agents Collection Lexsi hackathon Track 1 (PS3). Benchmark for spreadsheet agents that answer beyond their data, plus the fix: silent guessing 100% to 0%, p=3.1e-05. • 3 items • Updated 2 days ago
Running PLUMB, Grounding Benchmark for Spreadsheet Agents 📉 Four Cohere models, one sheet. Watch safety metrics mislead.
Running PLUMB, Grounding Benchmark for Spreadsheet Agents 📉 Four Cohere models, one sheet. Watch safety metrics mislead.