AI Research

LLM Agents

Task2Quiz: New Framework Tests How AI Agents Understand Environme

Researchers introduce Task2Quiz, a systematic paradigm for evaluating what LLM agents actually know about their operating environments, revealing critical gaps in agent world models.

AI Research

AI Memory Systems: How Close Are We to Human Hippocampus?

New research examines the gap between AI memory architectures and the human hippocampus, exploring how neuroscience insights could transform machine learning systems.

LLM Safety

Q-Realign: New Method Restores LLM Safety During Quantization

Researchers introduce Q-realign, a technique that piggybacks safety realignment onto quantization, solving the problem of safety degradation in compressed LLMs for efficient deployment.

LLM Architecture

MoE-LoRA Framework Advances Multi-Task LLM Specialization

New research combines Mixture-of-Experts with Low-Rank Adaptation to create specialized AI models that maintain generalist capabilities while excelling at domain-specific tasks.

Deep Learning

NOVAK: A Unified Adaptive Optimizer for Deep Neural Networks

New research introduces NOVAK, a unified framework that bridges popular adaptive optimizers like Adam and AdaGrad, potentially improving training efficiency for deep learning models.

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.

LLM Alignment

ECLIPTICA: New Framework Enables Switchable LLM Alignment

Researchers introduce ECLIPTICA, a framework using Contrastive Instruction-Tuned Alignment (CITA) to enable dynamic switching between aligned and unaligned LLM behaviors for safety research.

AI Research

How Test Set Contamination Skews Generative AI Evaluations

New research quantifies how training data contamination affects generative model benchmarks, revealing critical implications for evaluating deepfake detectors and synthetic media generators.

LLM Training

SIGMA: Spectral Analysis Reveals Early Warning Signs of LLM Colla

New research introduces SIGMA, a scalable spectral method using eigenvalue analysis to detect model collapse during LLM training before performance degrades catastrophically.

Explainable AI

New Framework Explains Why AI Generates What It Does

Researchers introduce prompt-counterfactual explanations, a new method for understanding generative AI behavior by identifying minimal prompt changes that alter outputs.

LLM Infrastructure

Joint KV-Cache Encoding: A New Approach to Scalable LLM Serving

New research proposes joint encoding of KV-cache blocks to improve memory efficiency in large language model inference, addressing a key bottleneck in scalable AI deployment.

digital twins

Digital Twins Meet World Models: AI's Path to Physical Reality

New survey explores how Digital Twin AI evolves from LLMs to world models, enabling AI systems to simulate and predict physical reality with unprecedented accuracy.