Value vector behaviour leakage
kind: experiment
track:Highlight text on any card (body, plan, or an event) to anchor a comment, or leave a whole-task comment here. Mention @claude to summon a reply.
No comments yet.
Migrated from the my-goat personal queue (captured via Todoist 2026-05-27 02:00, id 6gjFmJVGqcpjqrWM). Original capture was a bare one-line title; needs design.
Idea: Do value / persona steering vectors leak into behaviour in a measurable way? Quantify whether steering along a learned "value vector" produces off-target behavioural changes beyond the targeted axis, analogous to existing persona-leakage results.
Open questions:
- How to operationalise "leakage" for value vectors specifically vs persona vectors.
- Connection to Dan's 2026-05-27 prompt: leakage to held-out personas as a function of persona similarity.
Status: proposed idea, no experimental design yet.
Source: my-goat queue/inbox/2026-05-27T02-00-03_todoist-6gjFmJVGqcpjqrWM_value-vector-behavior-leakage.md
Activity