AI ethics evidence index Search
Source-linked daily record
Evidence record 6380 · automatically gathered

MetaSAEs: Joint Training with a Decomposability Penalty Produces More Atomic Sparse Autoencoder Latents

Sparse autoencoders (SAEs) are increasingly used for safety-relevant applications including alignment detection and model steering. These use cases require SAE latents to be as atomic as possible. Each latent should represent a single coherent concept drawn from a single underlying representational subspace. In practice, SAE latents blend representational subspaces together. A single feature can activate across semantically distinct contexts that share no true common representation, muddying an

Record details

Published: 3 April 2026
Source: arXiv
Category: Research
Topics: Safety & alignment
Retrieved: 14 July 2026

source-onlyevidence status

These records share source-supplied organisations, an exact publisher byline, automatic topics or regions. The reason is shown on every link; related does not mean supporting, agreeing with or verifying this record.

Shared topic: Safety & alignment

Large Language Models Align with the Human Brain during Creative Thinking

arXiv · 3 April 2026

Shared topic: Safety & alignment

Coupled Control, Structured Memory, and Verifiable Action in Agentic AI (SCRAT -- Stochastic Control with Retrieval and Auditable Trajectories): A Comparative Perspective from Squirrel Locomotion and Scatter-Hoarding

arXiv · 3 April 2026

Shared topic: Safety & alignment

Evaluating Artificial Intelligence Through a Christian Understanding of Human Flourishing

arXiv · 3 April 2026

Shared topic: Safety & alignment

Structural Rigidity and the 57-Token Predictive Window: A Physical Framework for Inference-Layer Governability in Large Language Models

arXiv · 4 April 2026

Shared topic: Safety & alignment

Automated Analysis of Global AI Safety Initiatives: A Taxonomy-Driven LLM Approach

arXiv · 4 April 2026

Shared topic: Safety & alignment

Disentangled Dual-Branch Graph Learning for Conversational Emotion Recognition

arXiv · 3 April 2026

How to cite this record

ethics.ai (3 April 2026), “MetaSAEs: Joint Training with a Decomposability Penalty Produces More Atomic Sparse Autoencoder Latents,” evidence record 6380, https://ethics.ai/record/6380 (originally published by arXiv).

JSON

Use and limitations

This page is a stable index and citation surface for a source record. ethics.ai did not author the underlying report and has not independently verified every claim. Automatic topics may be imperfect. For consequential use, quote and cite the original publisher.