Predicting Answer Correctness on TriviaQA (val)
0.904AUROCBaseline (Llama-3.1-8B)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Baseline (Llama-3.1-8B)Features=all UQ, Strategy=baseline2026.06 | 0.904 | — | — | |
| Gated Experts (Llama-3.1-8B)Features=all UQ, Strategy=latent gated2026.06 | 0.904 | — | — | |
| Selective Prediction (Llama-3.1-8B)Features=all UQ, Strategy=latent selective2026.06 | 0.904 | — | — | |
| Selective Prediction (Spectrum-Qwen3-14B)Features=all UQ, Strategy=latent selective2026.06 | 0.904 | — | — | |
| Gated Experts Oracle (Llama-3.1-8B)Features=all UQ, Strategy=oracle gated2026.06 | 0.903 | — | — | |
| Selective Prediction Oracle (Llama-3.1-8B)Features=all UQ, Strategy=oracle selective2026.06 | 0.903 | — | — | |
| Gated Experts (Spectrum-Qwen3-14B)Features=all UQ, Strategy=latent gated2026.06 | 0.903 | — | — | |
| Spectrum-Qwen3-14BFeatures=all UQ, Strategy=baseline2026.06 | 0.902 | — | — | |
| all UQBackbone=Spectrum-Qwen3-14B, Framework=latent selective2026.06 | 0.897 | — | — | |
| all UQBackbone=Spectrum-Qwen3-14B, Framework=latent gated2026.06 | 0.896 | — | — | |
| all UQBackbone=Spectrum-Qwen3-14B, Framework=baseline2026.06 | 0.895 | — | — | |
| all UQBackbone=Spectrum-Qwen3-14B, Framework=oracle gated2026.06 | 0.895 | — | — | |
| all UQBackbone=Spectrum-Qwen3-14B, Framework=oracle selective2026.06 | 0.895 | — | — | |
| all UQBackbone=Llama-3.1-8B, Framework=latent selective2026.06 | 0.888 | — | — | |
| SARBackbone=Spectrum-Qwen3-14B, Framework=latent selective2026.06 | 0.888 | — | — | |
| Selective Prediction (Spectrum-Llama-3.1-8B)Features=all UQ, Strategy=latent selective2026.06 | 0.887 | — | — | |
| all UQBackbone=Llama-3.1-8B, Framework=latent gated2026.06 | 0.887 | — | — | |
| SARBackbone=Spectrum-Qwen3-14B, Framework=latent gated2026.06 | 0.887 | — | — | |
| all UQBackbone=Llama-3.1-8B, Framework=baseline2026.06 | 0.886 | — | — | |
| all UQBackbone=Llama-3.1-8B, Framework=oracle gated2026.06 | 0.886 | — | — | |
| all UQBackbone=Llama-3.1-8B, Framework=oracle selective2026.06 | 0.886 | — | — | |
| CoCoABackbone=Spectrum-Qwen3-14B, Framework=latent gated2026.06 | 0.886 | — | — | |
| CoCoABackbone=Spectrum-Qwen3-14B, Framework=latent selective2026.06 | 0.886 | — | — | |
| SARBackbone=Spectrum-Qwen3-14B, Framework=baseline2026.06 | 0.886 | — | — | |
| SARBackbone=Spectrum-Qwen3-14B, Framework=oracle gated2026.06 | 0.886 | — | — | |
| SARBackbone=Spectrum-Qwen3-14B, Framework=oracle selective2026.06 | 0.886 | — | — | |
| Gated Experts (Spectrum-Llama-3.1-8B)Features=all UQ, Strategy=latent gated2026.06 | 0.885 | — | — | |
| Semantic entropyBackbone=Spectrum-Qwen3-14B, Framework=latent selective2026.06 | 0.885 | — | — | |
| Spectrum-Llama-3.1-8BFeatures=all UQ, Strategy=baseline2026.06 | 0.884 | — | — | |
| CoCoABackbone=Spectrum-Qwen3-14B, Framework=baseline2026.06 | 0.884 | — | — | |
| CoCoABackbone=Spectrum-Qwen3-14B, Framework=oracle gated2026.06 | 0.884 | — | — | |
| CoCoABackbone=Spectrum-Qwen3-14B, Framework=oracle selective2026.06 | 0.884 | — | — | |
| Semantic entropyBackbone=Spectrum-Qwen3-14B, Framework=latent gated2026.06 | 0.884 | — | — | |
| Semantic entropyBackbone=Spectrum-Qwen3-14B, Framework=oracle gated2026.06 | 0.882 | — | — | |
| Semantic entropyBackbone=Spectrum-Qwen3-14B, Framework=oracle selective2026.06 | 0.882 | — | — | |
| Semantic entropyBackbone=Spectrum-Qwen3-14B, Framework=baseline2026.06 | 0.881 | — | — | |
| all UQBackbone=Spectrum-Llama-3.1-8B, Framework=latent selective2026.06 | 0.88 | — | — | |
| Semantic energyBackbone=Spectrum-Qwen3-14B, Framework=latent selective2026.06 | 0.88 | — | — | |
| all UQBackbone=Spectrum-Llama-3.1-8B, Framework=latent gated2026.06 | 0.878 | — | — | |
| Semantic energyBackbone=Spectrum-Qwen3-14B, Framework=latent gated2026.06 | 0.878 | — | — | |
| all UQBackbone=Spectrum-Llama-3.1-8B, Framework=baseline2026.06 | 0.877 | — | — | |
| all UQBackbone=Spectrum-Llama-3.1-8B, Framework=oracle gated2026.06 | 0.877 | — | — | |
| all UQBackbone=Spectrum-Llama-3.1-8B, Framework=oracle selective2026.06 | 0.877 | — | — | |
| Semantic energyBackbone=Spectrum-Qwen3-14B, Framework=baseline2026.06 | 0.877 | — | — | |
| Semantic energyBackbone=Spectrum-Qwen3-14B, Framework=oracle gated2026.06 | 0.877 | — | — | |
| Semantic energyBackbone=Spectrum-Qwen3-14B, Framework=oracle selective2026.06 | 0.877 | — | — | |
| CoCoABackbone=Llama-3.1-8B, Framework=latent selective2026.06 | 0.875 | — | — | |
| CoCoABackbone=Llama-3.1-8B, Framework=baseline2026.06 | 0.874 | — | — | |
| CoCoABackbone=Llama-3.1-8B, Framework=latent gated2026.06 | 0.874 | — | — | |
| CoCoABackbone=Llama-3.1-8B, Framework=oracle gated2026.06 | 0.874 | — | — | |
| CoCoABackbone=Llama-3.1-8B, Framework=oracle selective2026.06 | 0.874 | — | — | |
| SARBackbone=Llama-3.1-8B, Framework=latent gated2026.06 | 0.874 | — | — | |
| SARBackbone=Llama-3.1-8B, Framework=latent selective2026.06 | 0.874 | — | — | |
| SARBackbone=Llama-3.1-8B, Framework=baseline2026.06 | 0.873 | — | — | |
| SARBackbone=Llama-3.1-8B, Framework=oracle gated2026.06 | 0.873 | — | — | |
| SARBackbone=Llama-3.1-8B, Framework=oracle selective2026.06 | 0.873 | — | — | |
| Selective Prediction (Qwen3-14B)Features=all UQ, Strategy=latent selective2026.06 | 0.872 | — | — | |
| all UQBackbone=Qwen3-14B, Framework=latent selective2026.06 | 0.872 | — | — | |
| Semantic entropyBackbone=Llama-3.1-8B, Framework=latent selective2026.06 | 0.871 | — | — | |
| Semantic entropyBackbone=Llama-3.1-8B, Framework=latent gated2026.06 | 0.87 | — | — | |
| Gated Experts (Llama-3.1-8B)Features=MI, Strategy=latent gated2026.06 | 0.869 | — | — | |
| Semantic entropyBackbone=Llama-3.1-8B, Framework=baseline2026.06 | 0.868 | — | — | |
| Semantic entropyBackbone=Llama-3.1-8B, Framework=oracle gated2026.06 | 0.868 | — | — | |
| Semantic entropyBackbone=Llama-3.1-8B, Framework=oracle selective2026.06 | 0.868 | — | — | |
| Semantic energyBackbone=Llama-3.1-8B, Framework=latent selective2026.06 | 0.867 | — | — | |
| Semantic entropyBackbone=Spectrum-Llama-3.1-8B, Framework=latent selective2026.06 | 0.867 | — | — | |
| SARBackbone=Spectrum-Llama-3.1-8B, Framework=latent selective2026.06 | 0.867 | — | — | |
| Semantic energyBackbone=Llama-3.1-8B, Framework=latent gated2026.06 | 0.866 | — | — | |
| Semantic energyBackbone=Llama-3.1-8B, Framework=baseline2026.06 | 0.865 | — | — | |
| Semantic energyBackbone=Llama-3.1-8B, Framework=oracle gated2026.06 | 0.865 | — | — | |
| Semantic energyBackbone=Llama-3.1-8B, Framework=oracle selective2026.06 | 0.865 | — | — | |
| SARBackbone=Spectrum-Llama-3.1-8B, Framework=latent gated2026.06 | 0.865 | — | — | |
| all UQBackbone=Qwen3-14B, Framework=latent gated2026.06 | 0.864 | — | — | |
| Semantic entropyBackbone=Spectrum-Llama-3.1-8B, Framework=latent gated2026.06 | 0.864 | — | — | |
| SARBackbone=Spectrum-Llama-3.1-8B, Framework=baseline2026.06 | 0.864 | — | — | |
| SARBackbone=Spectrum-Llama-3.1-8B, Framework=oracle gated2026.06 | 0.864 | — | — | |
| SARBackbone=Spectrum-Llama-3.1-8B, Framework=oracle selective2026.06 | 0.864 | — | — | |
| Gated Experts (Qwen3-14B)Features=all UQ, Strategy=latent gated2026.06 | 0.863 | — | — | |
| CoCoABackbone=Spectrum-Llama-3.1-8B, Framework=latent selective2026.06 | 0.863 | — | — | |
| Semantic energyBackbone=Spectrum-Llama-3.1-8B, Framework=latent selective2026.06 | 0.863 | — | — | |
| MIBackbone=Spectrum-Qwen3-14B, Framework=latent selective2026.06 | 0.863 | — | — | |
| SARBackbone=Qwen3-14B, Framework=latent selective2026.06 | 0.862 | — | — | |
| all UQBackbone=Qwen3-14B, Framework=baseline2026.06 | 0.861 | — | — | |
| all UQBackbone=Qwen3-14B, Framework=oracle gated2026.06 | 0.861 | — | — | |
| Semantic energyBackbone=Qwen3-14B, Framework=latent selective2026.06 | 0.861 | — | — | |
| CoCoABackbone=Spectrum-Llama-3.1-8B, Framework=latent gated2026.06 | 0.861 | — | — | |
| Semantic entropyBackbone=Spectrum-Llama-3.1-8B, Framework=oracle gated2026.06 | 0.861 | — | — | |
| MIBackbone=Spectrum-Qwen3-14B, Framework=latent gated2026.06 | 0.861 | — | — | |
| all UQBackbone=Qwen3-14B, Framework=oracle selective2026.06 | 0.86 | — | — | |
| CoCoABackbone=Spectrum-Llama-3.1-8B, Framework=oracle gated2026.06 | 0.86 | — | — | |
| Semantic energyBackbone=Spectrum-Llama-3.1-8B, Framework=latent gated2026.06 | 0.86 | — | — | |
| Semantic entropyBackbone=Spectrum-Llama-3.1-8B, Framework=baseline2026.06 | 0.86 | — | — | |
| Semantic entropyBackbone=Spectrum-Llama-3.1-8B, Framework=oracle selective2026.06 | 0.86 | — | — | |
| MSPBackbone=Spectrum-Qwen3-14B, Framework=latent selective2026.06 | 0.86 | — | — | |
| CoCoABackbone=Spectrum-Llama-3.1-8B, Framework=baseline2026.06 | 0.859 | — | — | |
| CoCoABackbone=Spectrum-Llama-3.1-8B, Framework=oracle selective2026.06 | 0.859 | — | — | |
| Semantic energyBackbone=Spectrum-Llama-3.1-8B, Framework=oracle gated2026.06 | 0.858 | — | — | |
| Semantic energyBackbone=Spectrum-Llama-3.1-8B, Framework=baseline2026.06 | 0.857 | — | — | |
| Semantic energyBackbone=Spectrum-Llama-3.1-8B, Framework=oracle selective2026.06 | 0.857 | — | — | |
| MSPBackbone=Spectrum-Qwen3-14B, Framework=latent gated2026.06 | 0.857 | — | — |