Download Epoch AI’s free dataset of notable AI models, chart training compute over time in Python, estimate how fast it grows, and count how many models cross the EU AI Act’s compute threshold.
Compute trends drive forecasts, export controls and legal thresholds. Reproducing them is a good first project for a coder interested in governance.
Output: A notebook with one chart, a growth-rate estimate and a short note.
Download the notable AI models dataset from Epoch AI’s data page (it is free, as a CSV). Open a Google Colab notebook and load it with pandas.
Plot training compute against publication date with a log scale on the y-axis. Label a few famous models so the chart tells a story.
Fit a straight line to log10(compute) for models since 2010 and convert the slope into “× per year” and a doubling time. Compare your answer with Epoch’s published estimate.
The EU AI Act presumes “systemic risk” for general-purpose models trained with more than 10²⁵ FLOP. Count how many models in the dataset exceed it, and when the first one did.
Share the chart, your estimate, and the biggest caveat (for example, missing or estimated compute values for some models).
All steps done
Share what you made, then try another project. The PDF version has room for notes.
Next: Estimate how much compute a famous model needed Next steps
Use the rule of thumb “training compute ≈ 6 × parameters × training tokens” to estimate the compute behind a model, then compare your answer with Epoch AI’s database.
Export controls target data-centre chips. But consumer hardware keeps getting better, and frontier-level performance reaches it within about a year. Update the analysis with the latest hardware.
Why do so many governance proposals focus on chips? Read one key paper and write a jargon-free explainer for someone outside tech.
Pick a known AI cluster from Epoch AI’s GPU cluster data, find it in free satellite imagery and document how it has grown over time.
Summarise how the EU sorts AI systems by risk (banned, high-risk, transparency duties, minimal) with one everyday example for each level, plus what changes for general-purpose models.
