LLM
Universal Latent Space Enables Zero-Shot LLM Routing
New research introduces a universal latent space approach for cost-efficient LLM routing, enabling zero-shot model selection without task-specific training data or expensive benchmarking.
Machine Learning
LLM Infrastructure
AIConfigurator Speeds Up LLM Serving Optimization Dramatically
New research introduces AIConfigurator, a system that dramatically accelerates configuration optimization for multi-framework LLM serving, enabling faster deployment of AI inference infrastructure.
LLM Safety
Global Subspace Projection: A New Approach to LLM Detoxification
Researchers propose a novel technique for removing toxic behaviors from large language models by projecting out malicious representations in the model's latent space.
transformer architecture
Transformer Architecture Explained: The Engine Behind Modern AI
A deep dive into the transformer architecture that powers everything from ChatGPT to AI video generators. Understanding attention mechanisms and why this design revolutionized machine learning.
AI Agents
Deep Agents: Solving Multi-Step AI Agent Failure Modes
AI agents often fail after several steps due to error compounding and context degradation. Deep Agents architecture introduces new mechanisms to maintain coherence across extended task execution.
LLM Quantization
FLRQ: Faster LLM Quantization via Low-Rank Matrix Sketching
New quantization method FLRQ achieves up to 2.5x faster compression of large language models while maintaining accuracy through flexible low-rank matrix approximation techniques.
embeddings
How AI Embeddings Power Everything From Search to Deepfakes
Embeddings transform words, images, and audio into mathematical vectors that AI uses to understand meaning. This core technology powers everything from search engines to deepfake detection systems.
LLM
Proactive Memory Extraction Advances LLM Agent Capabilities
New research proposes proactive memory extraction for LLM agents, moving beyond static summarization to enable more dynamic knowledge retention and recall in autonomous AI systems.
LLM unlearning
Dual-Granularity Data Synthesis Advances LLM Unlearning Methods
New research introduces domain-to-instance framework for generating synthetic data to help large language models selectively forget harmful knowledge while preserving useful capabilities.
LLM research
Agent Drift: Measuring LLM Behavior Decay in Multi-Agent Systems
New research quantifies how LLM agents degrade over extended interactions in multi-agent systems, revealing critical reliability challenges for production AI deployments.
Explainable AI
xDNN(ASP): Logic-Based Explainability for Neural Networks
New research combines deep neural networks with Answer Set Programming to generate human-readable explanations for AI decisions, advancing interpretability crucial for detection systems.
LLM research
Sparse Autoencoders Enable Fine-Grained Control of LLM Reasoning
New research demonstrates how Sparse Autoencoders can steer LLM reasoning processes, enabling precise control over chain-of-thought behavior without retraining models.