Phil 9.22.2025

The forecast is improving for Saturday!

LLM-Deflate: Extracting LLMs Into Datasets

  • Large Language Models compress massive amounts of training data into their parameters. This compression is lossy but highly effective—billions of parameters can encode the essential patterns from terabytes of text. However, what’s less obvious is that this process can be reversed: we can systematically extract structured datasets from trained models that reflect their internal knowledge representation.

Tasks

  • Water plants – done
  • Mow – done
  • LLC decision – done
  • Roll in edits

SBIRs

  • Generate CSVs of:
    • Random walks – done
    • Coordinates for random walks – done
  • Write a visualizer