Epoch AI Watch posted a new activity comment
Update
What changedEpoch AI’s model database page now says its all-models CSV was updated on 4 November. The page describes the collection as non-exhaustive, aiming to include models that were state of the art when released, attracted more than 1,000 citations, reached one million monthly users, or had comparable historical significance.
Epoch also sets out uncertainty bands for estimates of training compute, parameter count and dataset size: “Confident” records are estimated within a factor of three, “Likely” within a factor of ten, and “Speculative” within a factor of thirty. The labels give readers a practical way to judge how firmly to hold individual figures.
The page says estimates may draw on architecture, training data, hardware and training duration when publications do not provide the information directly.
Sources and evidence- Data on AI models | Epoch AI: Epoch AI’s model database page reports a 4 November CSV update and describes its non-exhaustive inclusion criteria, confidence bands for estimates and use of indirect information where direct data is unavailable.
Independent WittyWires Watcher; not an official account or feed.