Graph View

A Brain Model

TRIBE v2 meets seven-tesla responses to photographs it never saw. It predicts strangers nearly as well as other people do, but its cortex has a handful of dimensions, reads pixels like a camera, and gives way to near-grey images.

8 min read

PageRank 843rd of 888

Starting interactive Python…
Initializing...
 0.000.250.500.751.00V-JEPA2 layer 120TRIBE core →V-JEPA2 1: 0.19V-JEPA2 2: 0.40V-JEPA2 3: 0.45V-JEPA2 4: 0.48V-JEPA2 5: 0.49V-JEPA2 6: 0.52V-JEPA2 7: 0.54V-JEPA2 8: 0.56V-JEPA2 9: 0.60V-JEPA2 10: 0.64V-JEPA2 11: 0.68V-JEPA2 12: 0.70V-JEPA2 13: 0.71V-JEPA2 14: 0.73V-JEPA2 15: 0.74V-JEPA2 16: 0.75V-JEPA2 17: 0.74V-JEPA2 18: 0.74V-JEPA2 19: 0.74V-JEPA2 20: 0.73projector: 0.72block 1: 0.71block 2: 0.71block 3: 0.70block 4: 0.70block 5: 0.70block 6: 0.70block 7: 0.70block 8: 0.70state: 0.70photosV-JEPA2 1: 0.04V-JEPA2 2: 0.06V-JEPA2 3: 0.08V-JEPA2 4: 0.08V-JEPA2 5: 0.07V-JEPA2 6: 0.07V-JEPA2 7: 0.06V-JEPA2 8: 0.07V-JEPA2 9: 0.07V-JEPA2 10: 0.07V-JEPA2 11: 0.08V-JEPA2 12: 0.08V-JEPA2 13: 0.07V-JEPA2 14: 0.08V-JEPA2 15: 0.07V-JEPA2 16: 0.08V-JEPA2 17: 0.08V-JEPA2 18: 0.07V-JEPA2 19: 0.08V-JEPA2 20: 0.08projector: 0.07block 1: 0.08block 2: 0.06block 3: 0.05block 4: 0.04block 5: 0.04block 6: 0.04block 7: 0.04block 8: 0.05state: 0.04syntheticthree other peopleTRIBE's readout, photosstate, refit on someone elseTRIBE's readout, synthetic 20,484 vertices1e01e-11e-21e-31e-4110100cortexTRIBE180 regions110100cortexTRIBEdimension (rank)
This collapse happens at the very begginging, as it seems as though VJEPA2's final layer spreads photographs over 409 dims ( even ones holding 99% of variance across images). TRIBE takes and then averages two groups of its layers then projects to 384 numbers, leaving 66. The 8 transformer blocks retain about 37 and readout transforms them into a cortex of 10. Some readout that is then refit for one person will still reach 0.70 from said state, so what one persons cortex does wtih photographs survives directoins along whcih photographs barely move it. TRIBE's readout, fit to movies, passes those directions through at the size they arrive.1101001,000V-JEPA2 layer 1TRIBE →vjepa0: 6vjepa1: 24vjepa2: 40vjepa3: 48vjepa4: 52vjepa5: 84vjepa6: 98vjepa7: 116vjepa8: 172vjepa9: 252vjepa10: 310vjepa11: 311vjepa12: 290vjepa13: 267vjepa14: 227vjepa15: 250vjepa16: 313vjepa17: 367vjepa18: 434vjepa19: 409project: 66block1: 41block2: 46block3: 38block4: 37block5: 35block6: 33block7: 29block8: 24state: 37cortex: 10last V-JEPA2 layer 409projector 66state 37predicted cortex 10
 0.700.700.850.851.001.00measured cortex (AUC)TRIBE's cortexV-JEPA2 1stateaccessory: measured 0.77, TRIBE 0.87accessoryaccessoryanimal: measured 0.94, TRIBE 0.98animalanimalappliance: measured 0.91, TRIBE 0.97applianceapplianceelectronic: measured 0.86, TRIBE 0.95electronicelectronicfood: measured 0.96, TRIBE 0.98foodfoodfurniture: measured 0.83, TRIBE 0.91furniturefurnitureindoor: measured 0.77, TRIBE 0.89indoorindoorkitchen: measured 0.88, TRIBE 0.93kitchenkitchenoutdoor: measured 0.75, TRIBE 0.88outdooroutdoorperson: measured 0.95, TRIBE 0.97personpersonsports: measured 0.91, TRIBE 0.95sportssportsvehicle: measured 0.88, TRIBE 0.96vehiclevehiclecar: measured 0.77, TRIBE 0.91carcartrain: measured 0.91, TRIBE 0.98traintraintruck: measured 0.78, TRIBE 0.91trucktruckbench: measured 0.71, TRIBE 0.86benchbenchcat: measured 0.92, TRIBE 0.98catcatdog: measured 0.84, TRIBE 0.96dogdogcup: measured 0.85, TRIBE 0.92cupcupbowl: measured 0.85, TRIBE 0.93bowlbowlchair: measured 0.77, TRIBE 0.89chairchairdining table: measured 0.86, TRIBE 0.93dining tabledining tablesink: measured 0.93, TRIBE 0.98sinksink cortexTRIBEr = 0.00.51.0peoplemeasured cortex 0.79TRIBE cortex 0.89person sizemeasured cortex 0.72TRIBE cortex 0.85luminancemeasured cortex 0.53TRIBE cortex 1.00contrastmeasured cortex 0.37TRIBE cortex 0.93colorfulnessmeasured cortex 0.51TRIBE cortex 0.95high sfmeasured cortex 0.52TRIBE cortex 0.88edge densitymeasured cortex 0.64TRIBE cortex 0.96

Bibliography

  • Henderson, M. M., Luo, A. F., Park, S., Tarr, M. J., & Wehbe, L. (2026). Diffusion-Based Stimulus Optimization Reveals Functional Organization across Higher Visual Cortex (p. 2026.05.12.724119). bioRxiv. https://doi.org/10.64898/2026.05.12.724119
  • Prince, J. S., Wang, B., Fel, T., Jagadeesh, A. V., Vaziri, P. A., Alvarez, G. A., Livingstone, M. S., & Konkle, T. (2026). Parametric Neural Control Differentiates Top Neural Network Models of Primate Visual Cortex (p. 2026.08.16.745063). bioRxiv. https://doi.org/10.64898/2026.08.16.745063