AI News Hub Logo

AI News Hub

Anthropic researchers detail "model spec midtraining", which adds a stage between pretraining and fine-tuning to improve generalization from alignment training (Anthropic)

Techmeme
Techmeme

Anthropic: Anthropic researchers detail “model spec midtraining”, which adds a stage between pretraining and fine-tuning to improve generalization from alignment training  —  Sara Price2, Samuel Marks2,†, Jon Kutasov2,†  —  1Anthropic Fellows Program; 2Anthropic; †Equal advising