claim
active
claim:reward-shaping-can-amplify-as-well-as-block-a-behavior-unlike-data-filteringReward shaping can amplify as well as block a behavior, unlike data filtering
Safeguards analysis.
Source paper
extracted_from(2026) · Leon Bergen · Usha Bhalla · Sidharth Baskaran · Max Loeffler +13
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Fig. 10, Table 10.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Subtract λ times concept score from reward.
- SAE-based filtering and reward shaping show lower off-target effects but lower overall reductionfinding0.793Fig. 9.
- Playful and poetic amplification support.
- Abstract; central distinction.
- Fig. 8.
- Rewards are simply predictable stimuli (and aversive stimuli are, by definition, surprising)claim0.769Redefines reward and punishment in terms of predictability.
- Table 2, row 3, showing equivalence when prior preferences match rewards.
- Central question: does geometry in activation space causally determine behavior?