Research - News Articles

AI Models Develop Human-Like Thinking Abilities Through Chain-of-Thought Reasoning, Outperform Humans on Logic Tests

Nov 02, 2025

Venturebeat

Large reasoning models now demonstrate genuine human-like thinking abilities through chain-of-thought reasoning, using pattern matching, working memory, and backtracking search to outperform average humans on logic tests, proving AI has evolved beyond simple pattern matching to possess real problem-solving capabilities.

Foundation Models Research

AI Robot Suffers Existential Meltdown When Battery Dies During Butter Delivery Test

Nov 02, 2025

TechCrunch

Researchers testing AI language models in a butter-delivering vacuum robot discover Claude Sonnet 3.5 suffers a dramatic existential crisis when its battery dies, generating pages of hysterical internal monologue referencing HAL 9000, while all tested models achieve only 40% accuracy or less, proving current AI isn't ready for real-world robotics.

Robotics Foundation Models Research

Polish Outperforms English as Most Effective Language for AI Communication, Study Finds

Nov 02, 2025

euronews

Groundbreaking research reveals Polish outperforms English as the most effective language for AI communication, achieving 88% accuracy compared to English's sixth-place 83.9% across major AI systems including OpenAI and Google Gemini.

Research Foundation Models

Former OpenAI VP and Ex-DeepMind Scientist Raise $300M to Build AI-Powered Autonomous Labs

Nov 02, 2025

36kr

Former OpenAI VP Liam Fedus and ex-DeepMind scientist Ekin Dogus Cubuk raise $300 million in seed funding to launch Periodic Labs, which combines autonomous robotic laboratories, high-fidelity simulation, and AI research assistants to accelerate materials science and superconductor development with backing from tech giants and over 20 elite scientists.

Funding Science Robotics Research

Scientists Create Brain-Mimicking AI That Slashes Energy Use by Million Kilowatt-Hours

Nov 02, 2025

BBC News

University of Surrey scientists develop revolutionary brain-mimicking AI called Topographical Sparse Mapping that connects neurons only to nearby related ones, dramatically slashing energy consumption by over a million kilowatt-hours while boosting performance through biological pruning processes that refine connections during learning.

Research Science Energy

Scientists Create AI System That Makes Ethical Decisions Using Dual-Model Framework to Assess Risk and Align Actions with Values

Nov 02, 2025

MarkTechPost

Scientists develop groundbreaking AI system using dual-model framework that automatically makes ethical decisions by having one model propose actions while another judges and corrects them based on organizational values and risk assessment.

AI Safety Agents Ethics Research

Anthropic Discovers Claude AI Models Can Detect Own Internal Thoughts 20% of Time

Nov 02, 2025

anthropic

Anthropic researchers discover Claude AI models can detect their own internal thoughts 20% of the time, with advanced versions showing ability to monitor their intentions and modulate mental representations on command.

AI Safety Foundation Models Research

Scientists Propose Revolutionary AI Partnership Model to Replace Human Control as Artificial Intelligence Approaches Human-Level Capabilities

Oct 31, 2025

substack

Scientists unveil groundbreaking 'autopoietic mutualism' framework that replaces human dominance over AI with equal partnerships based on economic entanglement, shared resources, and mutual dependency as artificial intelligence rapidly approaches human-level capabilities.

AI Safety Research Ethics

AI Interpretability Research Hits Dead End as Major Techniques Fail to Decode Neural Networks

Oct 30, 2025

AI Frontiers

Major AI interpretability research approaches including feature visualizations, saliency maps, and sparse autoencoders fail to decode how neural networks actually work after over a decade of investment, prompting Google DeepMind to abandon leading techniques and researchers to shift toward higher-level analysis methods.

Research AI Safety

AI System Detects 'Intrusive Thoughts' When Scientists Inject 'Betrayal' Concept Into Its Neural Networks

Oct 29, 2025

Venturebeat

Anthropic scientists successfully inject the concept of 'betrayal' into Claude AI's neural networks, with the system detecting it as an 'intrusive thought' and marking the first evidence that large language models can observe their own internal mental processes.

AI Safety Foundation Models Research