Guide Labs debuts a new kind of interpretable LLM

The company open-sourced an 8 billion parameter LLM, Steerling-8B, trained with a new architecture designed to make its ...

Researchers baked 3x inference speedups directly into LLM weights — without speculative decoding

Researchers from the University of Maryland, Lawrence Livermore, Columbia and TogetherAI have developed a training technique that triples LLM inference speed without auxiliary models or infrastructure ...

Anthropic: Chinese AI firms created 24,000 fraudulent accounts for 'distillation attacks'

Anthropic says companies like DeepSeek are engaged in widespread fraud.
Security Boulevard

Model Inversion Attacks: Growing AI Business Risk

In an era where artificial intelligence (AI) and machine learning (ML) are driving unprecedented innovation and efficiency, a new class of cyber threats has emerged that puts sensitive data and entire ...