- How unreliable it is to choose a model based on leaderboards When it comes to choosing a model, I once even paid one yuan for Wenxin Yiyan. At the time, it was the first paid model in China, and I thought, could it really have something special? I paid to see whether it was actually any good, and in the end I got burned. This is the classic case: you see it perform very well on benchmarks, but in real applications it is especially weak. At the time, I just asked it to generate a webpage, and the final result was, well, hard to describe in one sentence. Later, when I tested Qwen, tested GLM, tested MiniMax, and tested DeepSeek, I ran into the same problem. In fact, over the past year, apart from those two companies, everyone had high scores but low ability. By
Purchase required to continue
This is a paid article. After signing in, your purchase will be unlocked automatically.
No comments yet. Be the first to share your thoughts.