Content A/B Testing on a Digital Brain Twin

I ran a LinkedIn post through Meta's TRIBE v2 — a model trained on 1,100 hours of brain scans from 700+ people — to predict how the human brain responds before anyone ever reads it.

The Brain Map

This is TRIBE v2's predicted fMRI activation for the final post — ~20,000 cortical vertices, mapped across four views:

Brain Region	Activation	What It Processes
Left Hemisphere	15.5%	Language, logic, analytical thinking
Right Hemisphere	14.3%	Creativity, emotion, spatial awareness
Top-Down View	15.2%	Motor planning, attention, focus
Front View	16.4%	Decision-making, social cognition

How It Works

TRIBE v2 processes text through a 3-stage pipeline:

1. Text → Speech (gTTS) with word-level timestamps (WhisperX)
2. Feature extraction via LLaMA 3.2 (text) + Wav2Vec-BERT (audio)
3. Brain mapping — Unified Transformer predicts ~20K cortical vertex activations per second

What is TRIBE v2?

Meta's trimodal brain encoder that predicts fMRI brain responses from video, audio, or text:

• Architecture: V-JEPA2 (video) + Wav2Vec-BERT (audio) + LLaMA 3.2 (text) → Unified Transformer
• Training data: 1,115 hours of fMRI from 720 subjects
• Resolution: ~70,000 voxels (70x improvement over v1)
• License: CC BY-NC 4.0

Links: Paper | GitHub | HuggingFace | Demo

Run It Yourself

Option 1: HuggingFace Space API (no GPU needed)

pip install gradio_client
python run_tribe_api.py

Calls Reino0ne/tribev2 Space — returns brain heatmaps.

Option 2: Google Colab (full 3D brain surface visualizations)

1. Upload tribe_demo.ipynb to Google Colab
2. Set runtime to T4 GPU
3. Accept LLaMA 3.2-3B license on HuggingFace
4. Add HF_TOKEN to Colab Secrets
5. Run all cells

Repo Structure

.
├── README.md
├── run_tribe_api.py          # Runs hooks via HF Space API
├── tribe_demo.ipynb          # Colab notebook with 3D brain maps
├── hooks/
│   ├── hook_a_storyteller.txt
│   ├── hook_b_provocative.txt
│   ├── hook_c_blunt.txt
│   └── hook_d_refined.txt    # Final version (posted)
└── results/
    ├── D_refined_brain.png   # Brain activation for final post
    ├── A_storyteller_brain.png
    ├── B_provocative_brain.png
    └── C_blunt_brain.png