FIELD NOTES TOPIC
tetrachoric-calibrated single-factor model
1 note, newest first.
When Does Combining Language Models Help? A Co-Failure Ceiling on Routing, Voting, and Mixture-of-Agents Across 67 Frontier Models
The text discusses research about using multiple language models (LLMs) to improve accuracy over single models. The study reveals that the gain from these systems is limited by a factor called 'beta', which represents…
