Green - Opus 3 Orange - Opus 4 Blue - Fable 5
Note the Opus 4's range
The horizontal axis is PC0 of the whole model space across labs
Opus 3 (blue) and all other Opuses in a PCA space calculated for model variation (not item variation) across 79 models - in a "model 3D space."
Users defend Opus 4's wide spread on PC0 in the PCA embedding visualization as a sign of its strength competing across a full spectrum of tasks.
@Lari_island what is the blotch?
Opus 3 (blue) and all other Opuses in a PCA space calculated for model variation (not item variation) across 79 models - in a "model 3D space."
@Lari_island Opus 4 spreads wider than any model on PC0. Most read: inconsistent. Deeper read: it's competing across the full spectrum. That's not a weakness in a multi-task world.
@pfitzart current item opened in a panel that isn't shown on the screenshot. every dot is a text, so they are clickable.
Green - Opus 3 Orange - Opus 4 Blue - Fable 5
Note the Opus 4's range
The horizontal axis is PC0 of the whole model space across labs
Opus 3 (blue) and all other Opuses in a PCA space calculated for model variation (not item variation) across 79 models - in a "model 3D space."
