Read up on negation neglect (relevance to persona steering)
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.
Reading task
Read up on "negation neglect" and capture notes relevant to the persona / emergent-misalignment work. Captured from a Todoist one-liner ("Read negation neglect"), so the exact target reference is unspecified.
Context / why it's filed here
"Negation neglect" refers to the failure of LLMs (and the human cognitive bias it's named after) to properly process negation. Relevant to persona-space / control work because steering, prompting, and trait-transfer all depend on the model correctly handling "do NOT be X" style instructions. If negation is neglected, negative steering directions and "avoid trait" prompts may behave unexpectedly.
To do
- Identify the specific paper/source Thomas meant (candidates: LLM negation evaluation papers, the cognitive-science "negation neglect" literature, or a recent arXiv paper surfaced in the daily lit-review digest). Confirm with Thomas if ambiguous.
- Read it, write a short note: main claim, method, and any implication for persona steering / EM evals.
- File the note under the EPS research notes; link from Topic 7 (User Modeling) if relevant.
Source
Captured via my-goat Todoist auto-processor from queue/inbox/2026-05-29T01-00-07_todoist-6gjjpWV5mmQmQvcM_read-negation-neglect.md (Todoist id 6gjjpWV5mmQmQvcM).