Skip to content

aogara-ds/ftm

partial_automa…
Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?
Code
This branch is 5 commits ahead, 5 commits behind epoch-research:main.

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
doc
 
 
ftm
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Full Takeoff Model

Work in progress! I don't recommend looking at the code right now. It needs to be cleaned up.

For now, you might be interested in the core file.

Instructions to execute the model:

  • Clone the github repo
  • Create a copy of the Task_based_model_inputs and set the parameters as you want, and make it public
  • Run python -m ftm.core.model -w YOUR_SHEET_URL to run the model with best_guess parameters
  • Run python -m ftm.analysis.exploration_analysis -w YOUR_SHEET_URL to run a comparison of the model with aggressive, best_guess and conservative parameter choices
    • You can run python -m ftm.analysis.exploration_analysis -w YOUR_SHEET_URL -t PARAMETER_NAME instead to produce a detailed comparison of what happens when you set said parameter at aggressive, best_guess and conservative, with all other parameters held at best_guess value
  • Run python -m ftm.analysis.sensitivity_analysis -w YOUR_SHEET_URL to see a high level comparison of what happens when you change each parameter between their aggressive, best_guess and conservative values, having all other parameters fixed at their best_guess value
    • The parameters will appear ordered by most sensitive to least sensitive. Concretely, the difference between the takeoff_length of the conservative and aggressive value, for a complex definition of takeoff_length
  • Run python -m ftm.analysis.mc_analysis -w YOUR_SHEET_URL to run a MC sampling
    • The aggressive, best guess and conservative values correspond to percentiles 5%, 50% and 95% of a distribution
    • You can adjust the correlation between parameters using the rank_correlation_between_buckets tab in the sheet
  • Run python -m ftm.analysis.timelines_report -w YOUR_SHEET_URL to run nine scenarios corresponding to the conservative, best_guess and aggresive choices, conditioned on short, best_guess and long AI timelines
    • The full_automation_requirement and flop_gap parameters are governed by the sheet Guess FLOP gap and timelines
  • Run python -m ftm.analysis.megareport -w YOUR_SHEET_URL to run the three previous analysis at once

About

Work in progress! I don't recommend looking at the code right now.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • HTML 96.3%
  • Python 2.2%
  • JavaScript 1.5%