This website is a supplement to the AI Takeoff Report by Tom Davidson, implemented by Epoch.
The best place to start learning about the report is by reading the short summary in Tom Davidson's report. The long summary discusses in more depth the results, while the rest of the report describes in detail the rationale behind the model.
In the model description section, you will find an interactive explanation of the Full Takeoff Model studied in the report. This will be useful to mathematically inclined readers who wish to understand the model deeply.
You can interact with the model, and set your own values for its input parameters, in the playground.
In the reports section, you will find three reports about the results we derive from the Full Takeoff Model:
- In the Monte Carlo report you will find the distributions of results that come from sampling the parameters according to Tom Davidson's beliefs.
- In the aggressive Monte Carlo report you will find the distributions of results that come from sampling the parameters according to Tom Davidson's beliefs but using an aggressive distribution for the amount of FLOP required to train an AGI.
- In the parameter importance report you will find a sensitivity analysis we run, highlighting the most important parameters for pinning down the results of the model
- In the timelines analysis you will find a comparative analysis of the model results conditional on beliefs that correspond to short, medium or long timelines until AGI.
The implementation of the model is available in our GitHub repository.
For bug reports and feature requests, please open a ticket in the repository or email firstname.lastname@example.org.
Jaime Sevilla and Eduardo Roldan were the main developers of the project, while Tom Davidson provided essential input through its development.
We thank Ege Erdil, Tom Adamczewski, Daniel Kokotajlo and the rest of the Epoch team for their feedback and support.