Open Distillation of Hereditary Traits

·Alignment Forum··

TL;DRJosh and Neel show that distillation from a teacher model to a base pretrained student model transfers some of the teacher model’s traits (such as displaying negative emotion in the Gemma Needs Help evals)On its own this is pretty unsurprising, but Josh and Neel additionally show that even filtering out all the prompts and rollouts where the trait is mentioned doesn’t generally prevent the trait transferIn this post, I show a simple way to replicate and study these phenomena without access ...

Read full article →

Related Articles

MIT's New Method Flags AI Models Trained on CASM Without Generating It
sdoering · Hacker News · 13h ago
How robust are natural language autoencoders to initialization?
michaelzhang · Alignment Forum · 4d ago
SFT Drives Gemini’s Safety Properties
Josh Engels · Alignment Forum · 1mo ago
Sequent: scale and automation for higher confidence in alignment
Geoffrey Irving · Alignment Forum · 1mo ago
A Mike's-Eye View of ARC's Research
Michael Winer · ARC · 1mo ago