AI research
A new position paper argues that a true science of AI requires studying training dynamics directly rather than relying on post-hoc fixes, with implications for understanding generative models and synthetic media systems.
interpretability
A new interpretability technique uses natural language autoencoders to translate opaque LLM internal activations into human-readable explanations, opening fresh approaches to AI transparency and synthetic content analysis.
AI safety
A new paper introduces multi-trait subspace steering to manipulate several behavioral dimensions in AI systems at once, offering a technical lens on alignment failure, misuse, and synthetic media safety.
LLM Research
A new paper introduces WASD, a method for finding neurons that are sufficient to explain and steer LLM behavior. The work adds technical insight into controllable generation and interpretable model editing.
LLM Research
Researchers propose a novel framework for visualizing and benchmarking factual hallucinations in large language models by analyzing internal neural activations and clustering patterns.
explainable-ai
New research introduces a dual-encoding approach to causal discovery, offering improved methods for understanding AI decision-making and model interpretability across complex systems.
LLM Research
Researchers propose a physics-inspired framework treating LLM token embeddings as discrete semantic states governed by Hamiltonian dynamics, offering new insights into transformer interpretability.
AI safety
New research shows AI models frequently omit key reasoning steps in their explanations, raising critical questions about whether we can trust AI transparency and the reliability of chain-of-thought prompting.
LLM Agents
New research introduces ABBEL, an architecture that constrains LLM agents to act through explicit belief states expressed in natural language, improving interpretability and decision-making in complex environments.
transformer-architecture
New research introduces a procedural task taxonomy to analyze why transformers struggle with compositional reasoning, offering insights for improving AI architecture design.
AI Alignment
New research proposes a cognitive architecture framework to address the 'black box' problem in AI systems, focusing on transparency, alignment, and interpretability through structured reasoning pathways.
synthetic-data
New research presents an interpretability-guided approach to generating synthetic emotional text data, addressing bias and quality concerns in AI-generated content through attention mechanism analysis and systematic evaluation.