Starling is model fine tuned using RLAIF (Reinforcement Learning from AI feedback), specifically using the berkeley-nest/Nectar dataset.
2.5 Not good
Pick which version of the model you base your review on.