Tag
42 articles
Nous Research introduces Contrastive Neuron Attribution (CNA), a method to steer LLM behavior without training or weight modification, preserving general capabilities.
Researchers explore OpenMythos, an open-source framework for building recurrent-depth transformers, focusing on MLA and GQA models and their parameter efficiency.
Spotify launches a new desktop research app as a research preview in over 20 markets, positioning itself against Google's NotebookLM.
Learn what Forward Deployed Engineers are, how they bridge the gap between AI research and real-world applications, and why this role is crucial for deploying AI technologies in business settings.
OpenAI co-founder Andrej Karpathy has joined Anthropic's pre-training team, highlighting the competitive race for top AI talent. Pre-training is a critical, compute-intensive phase in AI development that shapes a model's core capabilities.
Peter Steinberger is spending $1.3 million a month on OpenAI API usage to run 100 AI agents that automate coding, PR reviews, and bug detection, framing the expense as a research investment in the future of software development.
This explainer explores how AI video generation serves as a pathway to world models, the theoretical framework for creating general-purpose AI systems that understand and predict complex environments.
AI-generated research papers are becoming increasingly sophisticated, raising concerns about academic integrity and the credibility of scholarly databases. The phenomenon challenges current peer-review processes and citation metrics.
Google AI announced significant advancements in April 2026, including underwater video analysis and mobile AI integration that promise to revolutionize how we interact with technology.
This article explains how advanced AI systems like ChatGPT 5.5 Pro can perform PhD-level mathematical research, exploring the technical mechanisms behind AI reasoning and its implications for mathematical discovery.
Researchers have developed Talkie-1930, a 13-billion-parameter language model trained exclusively on pre-1931 English texts, to study historical reasoning and generalization.
Discord users gained unauthorized access to Anthropic's internal Mythos system, raising concerns about AI development security. The breach highlights the risks of community-driven exploration in sensitive AI environments.
