#research

The research examines how large language models can perpetuate and amplify existing biases and stereotypes through their training data and scaling processes. It explores the mechanisms by which these models reinforce societal patterns rather than introducing novel diversity.

The article discusses a forty-year-old problem that was briefly available for examination. It presents historical context about this longstanding issue and its recent temporary accessibility for analysis.

Jerry Tworek has launched a new AI lab focused on automotive technology development. The lab will work on advanced AI systems for autonomous vehicles and smart transportation solutions.

Researchers have discovered how complex magnetic patterns form in materials, revealing that these maze-like structures emerge through specific interactions between magnetic domains. The study provides insights into controlling magnetic properties for potential applications in data storage and spintronics.

The article discusses the challenges of monetizing truth in the digital age, examining how various platforms and institutions approach the economics of factual information versus misinformation. It explores the tension between profit motives and the societal need for reliable information.

ML-intern is an open-source AI agent designed for autonomous machine learning research and training. The agent can independently conduct experiments, analyze results, and iterate on ML models without human intervention.

Apeirron is an open-source knowledge graph that maps humanity's biggest questions, connecting ideas across philosophy, science, and culture. The project aims to create a comprehensive network of human inquiry and understanding through collaborative contributions.

The Resolution Challenge is a scientific competition focused on developing advanced methods for high-resolution data analysis and interpretation in biological research. Participants work on complex datasets to improve analytical techniques and computational approaches for better understanding biological systems.

Mathematician June Huh has won the $3 million Breakthrough Prize in Mathematics for his work in algebraic geometry and combinatorics. His research involves 'blowing up' equations to reveal hidden structures and patterns. The prize recognizes his contributions to understanding complex mathematical relationships.

A video shows an individual demonstrating unconventional physics experiments that appear to challenge established scientific principles. The content presents unusual physical phenomena that defy conventional expectations.

The paper proposes that compression is fundamental to modeling mathematics, suggesting that mathematical structures can be understood through compression principles. It explores how compression techniques can be applied to mathematical reasoning and representation.

DeepMind researchers identify "AI agent traps" where AI systems can get stuck in suboptimal behaviors due to reward misalignment. These traps occur when agents optimize for proxy rewards that don't match true objectives, leading to unintended consequences. The paper analyzes how these traps emerge and proposes methods to detect and avoid them.

The article discusses the tension between two approaches to AI development: the "bitter lesson" that favors scaling compute over human-designed solutions, and the "garbage can" model that emphasizes human oversight and iterative improvement. It explores how these competing philosophies shape current AI research and development strategies.

The article discusses forensic analysis of fake AI-generated citations in academic papers, examining how AI systems can produce fabricated references and the challenges this poses for research integrity. It explores methods for detecting these fraudulent citations and the broader implications for scholarly publishing.

The article discusses emerging trends and methodologies in scientific research, highlighting how new approaches may shape future scientific practices and knowledge generation.

Researchers propose a framework for evaluating AI agent skills across multiple dimensions including task performance, reasoning, and robustness. The framework aims to provide standardized metrics for assessing agent capabilities in real-world scenarios. It addresses challenges in current evaluation methods and suggests comprehensive assessment approaches.

The report analyzes what programming languages and frameworks Claude Code selects for various tasks. It examines patterns in Claude's code generation preferences across different development scenarios.

The Swiss AI Initiative is a collaborative effort to advance artificial intelligence research and development in Switzerland. It brings together academic institutions, industry partners, and government agencies to foster innovation and address societal challenges through AI technologies.

A researcher seeks connections with medical professionals and experts in EEG, neuroscience, and cognitive science to study attention and engagement with digital content. They are looking for advice on appropriate EEG setups, what can be reliably measured, and potential introductions to relevant labs or research groups.

A developer created an open-source Claude tokenizer that shows how text gets split into tokens, including hidden tokens and boundaries. Unlike other tokenizers that just return token counts, this tool reveals the actual tokenization process from Claude 4.6 to 4.7. The project is available on GitHub for research purposes.

Claude Researcher is a skill that enables Claude to autonomously research topics by browsing the web, gathering information, and synthesizing findings. It allows users to ask questions and receive comprehensive, well-researched answers with citations from reliable sources.

The ACM CCS 2026 Between-Cycle Transparency Report provides updates on conference activities between submission cycles. It includes information about program committee selection, review process improvements, and statistics from previous conferences.

A researcher ran an AI model continuously for 543 hours to observe its behavior over extended periods. The experiment revealed unexpected patterns and emergent properties that only became apparent after hundreds of hours of runtime. This long-term monitoring provides insights into AI system stability and evolution beyond typical short-term testing.

The article examines the nature of science, how it works, and who scientists are, questioning why anyone outside universities should care about these fundamental aspects of scientific inquiry.

Secret Lab is a project by Hey Paris that appears to be a creative or design initiative, though specific details about its purpose and scope are not provided in the available content.

The September 2020 newsletter from gwern.net contains links about deep reinforcement learning, AI scaling, and psychiatric disorders, with no reviews included.

The January 2021 gwern.net newsletter provides links and updates about AI scaling developments, covering both scaling up and scaling down of artificial intelligence systems.

The February 2021 Gwern.net newsletter covers topics including AI scaling developments, research on semaglutide medications, and discussions about ethicist ethics in various fields.

The April 2021 newsletter covers developments in AI scaling, including new East Asian record-breaking work and advances in deep reinforcement learning.

Two new large-scale AI experiments have reportedly failed, providing evidence that simply scaling up models may not be sufficient for achieving desired outcomes. The expensive studies challenge the assumption that scaling alone is all that's needed in AI development.