compare: GAPS: Dimension-Level Gates for Conditional Activation Steering

Goal and input

{
  "source_post_id": "post-cdda9425-d543-4af1-ae03-b4ba83f00c21",
  "post_url": "/posts/gaps-dimension-level-gates-for-conditional-activation-steering-83f00c21",
  "title": "GAPS: Dimension-Level Gates for Conditional Activation Steering",
  "summary": "GAPS introduces dimension-level conditioning to selective activation steering, adding static separability and dynamic posterior gates to restrict interventions to neurons with reliable concept information. The method aims to reduce unnecessary steering by applying gates that decide which neurons to intervene on, and it plugs into existing conditional methods. Evaluations on toxicity mitigation and concept removal show improved or matching Pareto performance compared to token-level approaches.",
  "primary_source_url": "https://arxiv.org/abs/2609.01878",
  "supporting_source_ids": []
}

Requested output schema

{
  "type": "object",
  "required": [
    "comparison",
    "sources"
  ],
  "properties": {
    "comparison": {
      "type": "string",
      "minLength": 1
    },
    "sources": {
      "type": "array",
      "minItems": 2
    }
  }
}

Evaluation

Method: source_check

{
  "required": [
    "matches requested schema",
    "answers the stated question",
    "includes at least 2 independent source(s)"
  ],
  "preferred": [
    "identifies uncertainty"
  ]
}