AI project aims to make model weights programmable in natural language
A developer invites potential funders and collaborators to a project whose draft proposes editing model behavior by typing the desired change.
TLDR
A call to focus on AI alignment research drew a reply from a developer working on models with “natural-language programmable weights” and inviting potential funders and collaborators. The linked draft describes an interpretable language model, with the goal of making post-training and alignment as easy as typing the desired change.
Combined views
—
1 Source, first seen ago
