METR Timelines Model

A probabilistic modeling tool for forecasting AGI timeline trajectories based on the paper by METR entitled "Measuring AI Ability to Complete Long Tasks".

This probabilistic approach accounts for uncertainties in starting capabilities, required end capabilities, and the rate of progress, providing a distribution of possible timelines rather than a single point estimate.

Installation

1. Clone the repository:

   git clone https://github.com/peterhurford/agi_timelines.git
   cd agi_timelines

2. Install dependencies using Poetry:

   # Install Poetry if you don't have it
   pip install poetry
   
   # Install dependencies
   poetry install

3. Run Jupyter notebook:

   poetry run jupyter notebook

Notes

• The model is based significantly on squigglepy, a Python library for Monte Carlo simulation.