Alignment

Our team is made up of researchers and professors from UCL, Oxford and Cambridge. Our products are powered by cutting-edge AI research.

Published on
February 8, 2024
Active Preference Learning for Large Language Models
ICML 2024LLM Fine-Tuning Alignment
We propose a practical acquisition function for prompt/completion pairs based on the predictive entropy of the language model and a measure of certainty of the implicit preference model optimized by DPO.

Active Preference Learning for Large Language Models