Scaling Law Prediction on ARC Challenge
0.0139MAENoLoss
Evaluation Results
| Method | Links | |
|---|---|---|
| NoLossAccess to ground truth accuracy=false, Pure forecasting=true, Model=Transformer (acc only)2026.01 | 0.0139 | |
| HistDiffAccess to ground truth accuracy=true, Model=Transformer (acc + histogram deltas)2026.01 | 0.0139 | |
| NEUNEUAccess to ground truth accuracy=false, Pure forecasting=true, Model=Transformer with CNN loss encoder, Parameters=20M2026.01 | 0.0146 | |
| AverageAccess to ground truth accuracy=false, Pure forecasting=true, Model=Transformer (acc + avg val prob)2026.01 | 0.02 | |
| LogisticAccess to ground truth accuracy=true, Type=Logistic scaling law fitted per task2026.01 | 0.0228 | |
| LC-PFNAccess to ground truth accuracy=false, Pure forecasting=true, Method=Transformer-based meta-learning2026.01 | 0.023 | |
| DiffProbeAccess to ground truth accuracy=true, Model=MLP predicting from histogram deltas2026.01 | 0.0236 |