Base Model Perplexity Regains Spotlight as Community Notes a Return to 2023
The community is debating a resurgence of ranking models by base pretraining perplexity, with @eliebakouch noting base log perplexity strongly predicts Epoch and OpenAI capability metrics, while others caution the landscape has changed since 2023.
2026-11-23 ~ 2026-11-23 · 2 related posts
- Pretraining perplexity alone can't rank base models anymore — stochasticchasm · 2026-11-23
- Base log perplexity is an extremely strong predictor of Epoch and OpenAI ECI, researchers note — stochasticchasm · 2026-11-23