Alignment + Rl + Safety

20.0

Build a library implementing Smooth Tchebysheff Scalarization for multi-objective offline RL. This helps researchers align models on multiple conflicting constraints without scalarization bias.

emergingimplementation gap

llminferencegovernancerlalignmentsafety

Signals (12)

tech review ai24d ago

Pareto-Optimal Offline Reinforcement Learning via Smooth Tchebysheff Scalarization

arXiv14d ago

How Do Language Models Process Ethical Instructions? Deliberation, Consistency, and Other-Recognition Across Four Models

arXiv7d ago

Beyond Social Pressure: Benchmarking Epistemic Attack in Large Language Models

arXiv8h ago

Correcting Suppressed Log-Probabilities in Language Models with Post-Transformer Adapters

arXiv1d ago

The Consciousness Cluster: Emergent preferences of Models that Claim to be Conscious

arXiv7d ago

Alignment + Rl + Safety

Signals (12)

The hardest question to answer about AI-fueled delusions

Introducing the Child Safety Blueprint

How to create “humble” AI

Responsible Scaling Policy

Reinforcement Learning via Value Gradient Flow

Why having “humans in the loop” in an AI war is an illusion

Pareto-Optimal Offline Reinforcement Learning via Smooth Tchebysheff Scalarization

How Do Language Models Process Ethical Instructions? Deliberation, Consistency, and Other-Recognition Across Four Models

Beyond Social Pressure: Benchmarking Epistemic Attack in Large Language Models

Correcting Suppressed Log-Probabilities in Language Models with Post-Transformer Adapters

The Consciousness Cluster: Emergent preferences of Models that Claim to be Conscious

Blind Refusal: Language Models Refuse to Help Users Evade Unjust, Absurd, and Illegitimate Rules