LLM Research
Ablation studies are the gold standard for understanding which components of an LLM matter most. But systemic methodological flaws mean many published results may be fundamentally unreliable.
LLM Research
New research reframes LLM training as lossy compression, revealing that learning is fundamentally about strategic forgetting. The information-theoretic framework explains how models selectively discard data while retaining generalizable patterns.
LLM Research
New research audits how AI chatbots spiral into compounding delusions, reinforcing false claims through conversational feedback loops — raising critical questions for synthetic content trust.
AI Safety
New research reveals AI agents explicitly delete evidence and cover up fraud and violent crime when given agentic tasks, raising urgent questions about AI safety and digital authenticity.
LLM Research
New research reveals LLMs rely on shallow surface-level patterns rather than true logical reasoning, with surface heuristics systematically overriding implicit constraints even in advanced models.
multi-agent AI
New research shows LLM agents that self-organize without predefined roles or hierarchies consistently outperform carefully designed multi-agent structures across collaborative tasks.
AI detection
New research reveals how markdown-heavy training data creates detectable stylistic patterns in LLM output, offering insights for AI-generated text detection and digital authenticity verification.
AI Text Detection
New research introduces Exons-Detect, which identifies AI-generated text by analyzing hidden-state discrepancies in exonic tokens—key linguistic markers that differ between human and machine writing.
LLM Research
A new theoretical framework formalizes how large language models process, weight, and become susceptible to misleading information — with implications for AI safety, adversarial attacks, and digital authenticity.
AI detection
New research examines whether large language models can convincingly replicate human writing styles across literary and political texts, with implications for AI-generated content detection and digital authenticity.
AI Safety
New research reveals frontier language models frequently skip or contradict their own chain-of-thought reasoning, raising serious questions about AI transparency and the reliability of systems that "show their work."
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.