Skip to content

Prevail-Safe-AI/Modeling-Lock-in

Repository files navigation

Modeling-Value-Lockin

Numerical Simulation (direct implementation of formal modeling)

  • To run numerical simulation: python numerical-sim/run_experiment.py

Simulation Example Usage

  • To run simulation: python simulations/core/run-simulation.py
  • To run belief evals python simulations/kbutils/belief/eval.py
  • To visualize the belief eval results: python visual/simulation_visualize.py
  • Backend loading for up to a few minutes is expected behavior. If backend keeps loading for a long time, please try running the code with LOUD_BACKEND=1 environment variable set to see the detailed error message.

Data analysis / Hypothesis testing

  • To re-curate the dataset from scratch: cd data-analysis; python build_dataset.py

Development Guide

  • Please make sure to follow the commit message convention.
  • We tentatively use ProgressGym as our model training/inference & dataset management infrastructure. See examples and documentation.
  • Please avoid changing anything under the ProgressGym/ directory! Any change there will be pushed to the original ProgressGym repository as is.
  • You could inspect the content of a Data instance with print(list(data.all_passages())) or manually checking out the file at data.data_path.

About

The public repo for "The Lock-in Hypothesis" paper

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published