Hoffmann et al. (2022) attribute to Kaplan et al. (2020) the conclusion that large models should not be trained to their lowest possible loss to be compute-optimal; the next sentence says they reach the same conclusion but recommend many more training tokens.
One recorded claim. The claim panel gives its qualification, the source passages with their place in each source, and how it was checked.
What it connects
Training Compute-Optimal Large Language Models → attributes to → Scaling Laws for Neural Language Models
Where it appears
No collection page holds this claim. It is listed with its records in the explorer.