Phil 6.15.20

The nice thing about riding big distances is that even though nothing has changed, everything is different and better for a while


GPT-2 Agents

  • Try to make a adjacency matrix from the DB. That may work, but it sure doesn’t generate anything human-readable. Need to roll my own
  • After accidentally blowing away my database (Yay, backups!), I’m reading in the network. This actually looks really good. I’d not an 8×8 grid, but the system found 63 nodes, and you can see that many adjacent nodes are connected:


  • You can also see that common moves, such as e2 and d2 are in the center and well connected.
  • Hand-rolled an adjacency matrix using pandas.DataFrame and exported to Excel. I have to think about what this means now. I think that it’s clear that moves tend to be nearby. I’m clearly not setting something right, because I don’t have the weight of the edge between nodes (I’m currently using the number of times a node was visited). Now I need to figure out how to use this:



  • Continue with revisions
  • After trying the pipeline image, (very small text and pix!), I’m going to try a new, more vertical layout. I think this is a little better. It’s a lot more legible:

Enter a caption

  • Ok, back to writing actual text

Good chat with Aaron about the Conspiracy as mode collapse experiments


  • Sy’s Presentation