List your company
ResearchFormerly WizWand

Linguistic Diversity Analysis on Real-world vs. Simulated User Conversations

7.65EntropyKimi-K2-Instruct
6.07966.48736.8957.3027Dec 23, 2025
Updated 8mo ago

Evaluation results

MethodLinks
Kimi-K2-InstructMode=zero-shot2025.127.650.4160.851
GPT-4oMode=zero-shot2025.127.120.4080.837
Llama-3.3-70B-InstructMode=zero-shot2025.127.020.2680.651
Real UsersType=Human Baseline2025.126.60.4830.861
UserSim-160K-it2Training Data Size=160K, Training Stage=iteration 22025.126.520.4390.821
UserSim-160K-it3Training Data Size=160K, Training Stage=iteration 32025.126.510.4610.841
UserSim-160K-it1Training Data Size=160K, Training Stage=iteration 12025.126.380.4710.839
UserSim-20K-baseTraining Data Size=20K, Training Stage=base2025.126.230.4220.782
UserSim-160K-baseTraining Data Size=160K, Training Stage=base2025.126.180.3820.728
UserSim-80K-baseTraining Data Size=80K, Training Stage=base2025.126.150.4090.759
UserSim-10K-baseTraining Data Size=10K, Training Stage=base2025.126.140.4060.756
UserSim-40K-baseTraining Data Size=40K, Training Stage=base2025.126.140.430.782