TAO: Using test-time compute to train efficient LLMs without labeled data
www.databricks.com
LIFT fine-tunes LLMs without labels using reinforcement learning, boosting performance on enterprise tasks.