LURE: Alignment Evaluations to Reduce Evaluation Awareness — LessWrong
TLDR: * Frontier models can detect when they're being evaluated and change their behavior, which risks compromising safety benchmarks. * We introdu…
America Forever Bytes
alignment · AFBytes
Other
alignment
Recent coverage
[2511.16992] FIRM: Federated In-client Regularized Multi-objective Alignment for Large Language Models
Abstract page for arXiv paper 2511.16992: FIRM: Federated In-client Regularized Multi-objective Alignment for Large Language Models
[2605.30808] Differentially Private Preference Data Synthesis for Large Language Model Alignment
Abstract page for arXiv paper 2605.30808: Differentially Private Preference Data Synthesis for Large Language Model Alignment
Why tuning fails: The AI has no self — LessWrong
---------------------------------------- …
How much should we worry about secretly loyal AIs? — LessWrong
Defenders have structural advantages but there’s work to be done
[2605.29365] Casual as an Anchor: Resolving Supervision Misalignment in Formality Transfer Dataset
Abstract page for arXiv paper 2605.29365: Casual as an Anchor: Resolving Supervision Misalignment in Formality Transfer Dataset
[2605.29442] How Coding Agents Fail Their Users: A Large-Scale Analysis of Developer-Agent Misalignment in 20,574 Real-World Sessions
Abstract page for arXiv paper 2605.29442: How Coding Agents Fail Their Users: A Large-Scale Analysis of Developer-Agent Misalignment in 20,574 Real-World Sessio...
[2605.29458] Adaptive Interviewing for Persona Simulation in LLMs: Evidence-Grounded Reasoning Improves Decision Alignment
Abstract page for arXiv paper 2605.29458: Adaptive Interviewing for Persona Simulation in LLMs: Evidence-Grounded Reasoning Improves Decision Alignment
[2605.28188] Framing Matters: Addressing Framing Sensitivity in Decision-Making through Behaviorally-Grounded Value Alignment
Abstract page for arXiv paper 2605.28188: Framing Matters: Addressing Framing Sensitivity in Decision-Making through Behaviorally-Grounded Value Alignment
[2605.28597] Position: Retire the "Positive Backdoor" Label -- Secret Alignment Requires Strict and Systematic Evaluation
Abstract page for arXiv paper 2605.28597: Position: Retire the "Positive Backdoor" Label -- Secret Alignment Requires Strict and Systematic Evaluation
[2605.27969] Boundary Suppression Asymmetry in Post-trained Assistants: Over-expansion as a Controllability Cost
Abstract page for arXiv paper 2605.27969: Boundary Suppression Asymmetry in Post-trained Assistants: Over-expansion as a Controllability Cost