Error Prediction on AmbigQA (val)
69.8PRRSpectrum-Qwen3-14B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Spectrum-Qwen3-14BFeatures=all UQ, Strategy=latent gated2026.06 | 69.8 | — | |
| Llama-3.1-8BFeatures=all UQ, Strategy=baseline2026.06 | 69.6 | — | |
| Spectrum-Qwen3-14BFeatures=all UQ, Strategy=baseline2026.06 | 69.6 | — | |
| Llama-3.1-8BFeatures=all UQ, Strategy=latent gated2026.06 | 69.5 | — | |
| Spectrum-Qwen3-14BFeatures=all UQ, Strategy=latent selective2026.06 | 68.8 | — | |
| Llama-3.1-8BFeatures=SAR, Strategy=latent gated2026.06 | 68.5 | — | |
| Llama-3.1-8BFeatures=SAR, Strategy=baseline2026.06 | 68.3 | — | |
| Llama-3.1-8BFeatures=all UQ, Strategy=latent selective2026.06 | 68.1 | — | |
| Llama-3.1-8BFeatures=CoCoA, Strategy=latent gated2026.06 | 65.1 | — | |
| oracle gatedModel=Llama-3.1-8B, Features=all UQ2026.06 | 65.1 | — | |
| Llama-3.1-8BFeatures=CoCoA, Strategy=baseline2026.06 | 64.4 | — | |
| latent gatedModel=Llama-3.1-8B, Features=all UQ2026.06 | 63.3 | — | |
| baselineModel=Llama-3.1-8B, Features=all UQ2026.06 | 63.2 | — | |
| oracle selectiveModel=Llama-3.1-8B, Features=all UQ2026.06 | 62.7 | — | |
| latent selectiveModel=Llama-3.1-8B, Features=all UQ2026.06 | 62 | — | |
| A2Search-7BFeatures=all UQ, Strategy=latent selective2026.06 | 61.6 | — | |
| A2Search-7BFeatures=all UQ, Strategy=latent gated2026.06 | 61.4 | — | |
| Spectrum-Llama-3.1-8BFeatures=all UQ, Strategy=latent selective2026.06 | 60.7 | — | |
| Spectrum-Llama-3.1-8BFeatures=all UQ, Strategy=latent gated2026.06 | 60.5 | — | |
| Spectrum-Llama-3.1-8BFeatures=all UQ, Strategy=baseline2026.06 | 60.2 | — | |
| A2Search-7BFeatures=all UQ, Strategy=baseline2026.06 | 59.3 | — | |
| Qwen3-14BFeatures=all UQ, Strategy=latent gated2026.06 | 58.4 | — | |
| latent gatedModel=A^2Search-7B, Features=all UQ2026.06 | 58 | — | |
| latent selectiveModel=A^2Search-7B, Features=all UQ2026.06 | 58 | — | |
| Qwen3-14BFeatures=all UQ, Strategy=baseline2026.06 | 57.9 | — | |
| oracle gatedModel=Qwen3-14B, Features=all UQ2026.06 | 57.5 | — | |
| oracle gatedFeatures=all UQ, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 57.5 | — | |
| oracle gatedModel=A^2Search-7B, Features=all UQ2026.06 | 57.3 | — | |
| latent gatedModel=Qwen3-14B, Features=all UQ2026.06 | 57.1 | — | |
| latent gatedFeatures=all UQ, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 57.1 | — | |
| oracle gatedModel=Spectrum-Qwen3-14B, Features=all UQ2026.06 | 56.5 | — | |
| Qwen3-14BFeatures=all UQ, Strategy=latent selective2026.06 | 56.4 | — | |
| latent selectiveModel=Qwen3-14B, Features=all UQ2026.06 | 56.2 | — | |
| latent selectiveFeatures=all UQ, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 56.2 | — | |
| baselineModel=Spectrum-Qwen3-14B, Features=all UQ2026.06 | 56 | — | |
| latent gatedModel=Spectrum-Qwen3-14B, Features=all UQ2026.06 | 56 | — | |
| latent selectiveModel=Spectrum-Qwen3-14B, Features=all UQ2026.06 | 55.6 | — | |
| baselineModel=Qwen3-14B, Features=all UQ2026.06 | 55.5 | — | |
| baselineFeatures=all UQ, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 55.5 | — | |
| latent gatedFeatures=CoCoA, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 55.3 | — | |
| oracle gatedFeatures=CoCoA, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 55.3 | — | |
| oracle selectiveModel=A^2Search-7B, Features=all UQ2026.06 | 54.5 | — | |
| baselineModel=A^2Search-7B, Features=all UQ2026.06 | 54.3 | — | |
| oracle gatedFeatures=SAR, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 54.3 | — | |
| latent gatedFeatures=SAR, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 53.9 | — | |
| baselineFeatures=CoCoA, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 53.8 | — | |
| oracle selectiveModel=Qwen3-14B, Features=all UQ2026.06 | 53.7 | — | |
| oracle selectiveFeatures=all UQ, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 53.7 | — | |
| oracle selectiveModel=Spectrum-Qwen3-14B, Features=all UQ2026.06 | 53.6 | — | |
| oracle selectiveFeatures=SAR, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 52.8 | — | |
| oracle selectiveFeatures=CoCoA, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 52.6 | — | |
| latent selectiveFeatures=SAR, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 52.6 | — | |
| baselineFeatures=SAR, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 52.5 | — | |
| latent selectiveFeatures=CoCoA, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 52.4 | — | |
| latent gatedFeatures=Sem. energy, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 52 | — | |
| latent selectiveFeatures=Sem. energy, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 52 | — | |
| latent gatedFeatures=Sem. entropy, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 51.3 | — | |
| latent gatedFeatures=MSP, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 51.2 | — | |
| latent selectiveFeatures=Sem. entropy, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 50.8 | — | |
| oracle gatedFeatures=MSP, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 50.4 | — | |
| latent selectiveFeatures=MSP, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 49.2 | — | |
| oracle gatedModel=Spectrum-Llama-3.1-8B, Features=all UQ2026.06 | 48.3 | — | |
| baselineFeatures=MSP, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 48.3 | — | |
| oracle selectiveFeatures=MSP, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 47.8 | — | |
| oracle gatedFeatures=Sem. entropy, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 47.1 | — | |
| latent gatedModel=Spectrum-Llama-3.1-8B, Features=all UQ2026.06 | 46.2 | — | |
| baselineModel=Spectrum-Llama-3.1-8B, Features=all UQ2026.06 | 46.1 | — | |
| oracle gatedFeatures=Sem. energy, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 46 | — | |
| latent selectiveModel=Spectrum-Llama-3.1-8B, Features=all UQ2026.06 | 45.7 | — | |
| oracle selectiveModel=Spectrum-Llama-3.1-8B, Features=all UQ2026.06 | 45 | — | |
| oracle selectiveFeatures=Sem. entropy, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 44.6 | — | |
| oracle selectiveFeatures=Sem. energy, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 43.2 | — | |
| baselineFeatures=Sem. entropy, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 42.1 | — | |
| baselineFeatures=Sem. energy, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 39.2 | — | |
| oracle gatedModel=Llama-3.1-8B, Features=MI2026.06 | 19.9 | — | |
| baselineModel=Llama-3.1-8B, Features=MI2026.06 | 18.9 | — | |
| latent gatedModel=Llama-3.1-8B, Features=MI2026.06 | 18.9 | — | |
| oracle selectiveModel=Llama-3.1-8B, Features=MI2026.06 | 18.2 | — | |
| latent selectiveModel=Llama-3.1-8B, Features=MI2026.06 | 18.1 | — | |
| latent gatedModel=Qwen3-14B, Features=MI2026.06 | 7.8 | — | |
| latent selectiveModel=Qwen3-14B, Features=MI2026.06 | 7.8 | — | |
| latent gatedFeatures=MI, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 7.8 | — | |
| latent selectiveFeatures=MI, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 7.8 | — | |
| oracle gatedModel=Qwen3-14B, Features=MI2026.06 | 5.8 | — | |
| oracle selectiveModel=Qwen3-14B, Features=MI2026.06 | 5.8 | — | |
| oracle gatedFeatures=MI, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 5.8 | — | |
| oracle selectiveFeatures=MI, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 5.8 | — | |
| Qwen3-14BFeatures=MI, Strategy=latent gated2026.06 | 3 | — | |
| baselineModel=Qwen3-14B, Features=MI2026.06 | 1.4 | — | |
| baselineFeatures=MI, Backbone=Qwen3-14B, Supervision=AlignScore2026.06 | 1.4 | — | |
| A^2Search-7B + all UQmodel=A^2Search-7B, features=all UQ, setting=baseline2026.06 | — | 75.7 | |
| A^2Search-7B + all UQmodel=A^2Search-7B, features=all UQ, setting=latent gated2026.06 | — | 76.7 | |
| A^2Search-7B + MImodel=A^2Search-7B, features=MI, setting=latent gated2026.06 | — | 55.5 | |
| A^2Search-7B + MImodel=A^2Search-7B, features=MI, setting=latent selective2026.06 | — | 55.5 | |
| Llama-3.1-8B + all UQmodel=Llama-3.1-8B, features=all UQ, setting=baseline2026.06 | — | 79.9 | |
| Llama-3.1-8B + all UQmodel=Llama-3.1-8B, features=all UQ, setting=latent gated2026.06 | — | 79.8 | |
| Llama-3.1-8B + all UQmodel=Llama-3.1-8B, features=all UQ, setting=oracle gated2026.06 | — | 79.9 | |
| Llama-3.1-8B + all UQmodel=Llama-3.1-8B, features=all UQ, setting=latent selective2026.06 | — | 79.7 | |
| Llama-3.1-8B + all UQmodel=Llama-3.1-8B, features=all UQ, setting=oracle selective2026.06 | — | 79.7 | |
| Llama-3.1-8B + CoCoAmodel=Llama-3.1-8B, features=CoCoA, setting=baseline2026.06 | — | 76.8 |