Skip to content
TheChatArchiveThe Chat Archive
Coherecommand-a-03-2025· CohereHugging Face LMArena Human Preference 140K

In each plot, the gray dots represent data points $(Xi,Yi)$, the dotted line represents…

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with Qwenqwen-max-2025-01-25’s answer.
  1. User

    In each plot, the gray dots represent data points (Xi,Yi)(X_i,Y_i), the dotted line represents the output of a quadratic regression algorithm, \cAquad\cA_{\textnormal{quad}}, and the solid line represents the output of a spline regression algorithm \cAspline\cA_{\textnormal{spline}}. The left plot shows the fitted models on the full data set ((Xj,Yj))j∈[n]((X_j,Y_j))_{j\in[n]}, while the middle and right plots show the fitted models trained on ((Xj,Yj))j∈[n]\{i}((X_j,Y_j))_{j\in[n]\backslash\{i\}}, where (Xi,Yi)(X_i,Y_i), marked by a yellow cross, has been removed. A stable algorithm is one for which the removal of a typical data point would not cause a large change in the predicted values returned by the fitted model (as in Definition~\ref{def:alg_stab_eps_delta}). In this figure, we can see that \cAquad\cA_{\textnormal{quad}} satisfies this property, but \cAspline\cA_{\textnormal{spline}} does not.


    Make this caption more efficient

  2. Cohere Assistant

    The plots compare the stability of quadratic regression (\cAquad\cA_{\textnormal{quad}}) and spline regression (\cAspline\cA_{\textnormal{spline}}) algorithms. The left plot shows fitted models on the full dataset, while the middle and right plots show models trained with one point (marked by a yellow cross) removed. \cAquad\cA_{\textnormal{quad}} exhibits stability, as removing a point causes minimal change in predictions, whereas \cAspline\cA_{\textnormal{spline}} shows significant variability, indicating instability.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.