Generative AI
Generative AI as Threshold Logic in High Dimensions
A new arXiv paper reframes generative AI as threshold logic operating in high-dimensional space, offering foundational insights into how neural networks produce synthetic content.
research
LLM Systems
Decision-Centric Design: A New Framework for LLM Systems
A new research paper proposes decision-centric design for LLM systems, shifting focus from model accuracy to downstream decision quality — with implications for how AI pipelines are architected.
AI Safety
Why Aligned AI Systems Remain Persistently Vulnerable
New research examines why safety alignment in large AI models remains fundamentally fragile, with implications for content guardrails meant to prevent deepfake and synthetic media generation.
model-compression
OneComp: One-Line Model Compression for Generative AI
A new framework called OneComp promises to compress generative AI models with a single line of code, potentially making diffusion and video generation models far more deployable at the edge.
AI Security
Survey Maps AI Security Landscape for Foundation Models
A new comprehensive survey provides a unified framework for understanding AI security threats across foundation models, covering adversarial attacks, deepfake generation, synthetic media detection, and content authenticity challenges.
AI Detection
Why AI Text Detectors Fail: Explainable AI Reveals Hidden Flaws
New research uses explainable AI techniques to reveal why AI-generated text detectors fail in practice despite strong benchmark scores, exposing critical shortcomings in current detection approaches.
LLM Inference
KV Cache Optimization: Key to Scalable LLM Inference
A comprehensive survey explores KV cache optimization strategies—from quantization to eviction policies—that make large language model inference faster, cheaper, and more scalable across generative AI applications.
adversarial attacks
Neural Uncertainty Principle Links Adversarial Attacks to LLM Hal
A new theoretical framework unifies adversarial vulnerability in neural networks with LLM hallucination, proposing that both arise from a fundamental uncertainty trade-off in learned representations.
AI Safety
Subspace Steering Exposes Risks in Human-AI Behavior
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 Bias
Name Swaps Expose Hidden Bias in LLM Judgments
A new paper shows that changing only names in prompts can flip LLM verdicts, revealing systematic bias through intervention consistency tests. The findings matter for AI moderation, authenticity review, and automated decision systems.
AI Video
Script-to-Slide Grounding Advances AI Video Creation
A new paper introduces script-to-slide grounding for automatic instructional video generation, linking script sentences to slide objects so systems can produce more structured, context-aware educational videos.
Speech Deepfakes
Codec-Aware Model Targets Speech Deepfake Detection
A new paper proposes quantizer-aware hierarchical neural codec modeling for speech deepfake detection, targeting artifacts introduced by modern neural audio codecs used in synthetic speech pipelines.