SILICON decomposes LLM annotation measurement error into guideline, baseline, prompt, and model sources, applies targeted fixes validated on nine tasks, and enables open-weight backups plus routing for long-term reproducibility.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2024 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
To Err Is Human; To Annotate, SILICON? Toward Robust Reproducibility in LLM Annotation
SILICON decomposes LLM annotation measurement error into guideline, baseline, prompt, and model sources, applies targeted fixes validated on nine tasks, and enables open-weight backups plus routing for long-term reproducibility.