Research map / Learning theory and algorithms / Prediction and decision losses
[Re] Value Alignment Verification
Siba Smarak Panigrahi, Sohan Patnaik
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use SCOT-selected trajectory states instead of critical-state queries | Black-box action-query alignment tests · Random grid navigation domains | sample complexity | Improved |
| Use SCOT instead of CS | Black-box action-query alignment tests · Random grid navigation domains | accuracy | Improved |
| Use ARP-bb instead of SCOT | Black-box action-query alignment tests · Random grid navigation domains | accuracy | Mixed |
| Use ARP-pref or ARP-bb rather than other action-query heuristics | Value alignment verification · Random grid navigation domains | sample efficiency | Improved |
| Use ARP-w or ARP-pref rather than approximate action-query tests | Value alignment verification · Random grid navigation domains | accuracy | Improved |
| Use ARP-bb rather than SCOT | Black-box action-query alignment tests · Random grid navigation domains | sample complexity | Improved |
| Use SCOT rather than CS | Black-box action-query alignment tests · Random grid navigation domains | sample complexity | Improved |
| Increase the number of elicited synthetic human preferences | Approximate implicit-value alignment tests · Autonomous driving synthetic human preferences | verification accuracy | Improved |
| Increase the preference-query budget beyond 50 | Approximate implicit-value alignment tests · Autonomous driving synthetic human preferences | verification accuracy | No clear effect |
| Use elicited real-human preferences to construct an alignment test | Approximate implicit-value alignment verification · Autonomous driving real human preference pilot | verification accuracy: 72% | Improved |
Closest papers by meaning
- FormalAlign: Automated Alignment Evaluation for Autoformalization (ICLR 2025 · similarity 0.81)
- Conformal Alignment: Knowing When to Trust Foundation Models with Guarantees (NeurIPS 2024 · similarity 0.78)
- Perplexity-aware Correction for Robust Alignment with Noisy Preferences (NeurIPS 2024 · similarity 0.77)
- MAP: Multi-Human-Value Alignment Palette (ICLR 2025 · similarity 0.77)
- Inference-Time Personalized Alignment with a Few User Preference Queries (NeurIPS 2025 · similarity 0.77)
- Scalable Valuation of Human Feedback through Provably Robust Model Alignment (NeurIPS 2025 · similarity 0.77)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
