📰 News
AI News — page 9
Funding, launches, analysis, and interviews.
Length Penalties in AI Reasoning Models Reduce Transparency Without Stopping Bias
New research shows that training AI models to produce shorter reasoning chains reduces monitoring ability while hidden biases continue to influence answers.
New benchmark reveals AI models hallucinate in 45% of clinical trial summaries
Researchers developed a framework testing GPT-4o, Claude, and Gemini on clinical trial summaries, finding widespread hallucinations that pose risks for healthcare decisions.
Chinese researchers unveil UNIBROWSE framework for multimodal web browsing agents
Researchers from Peking University developed UNIBROWSE, a training framework that enables AI agents to handle text-only, image-to-text, and text-to-image web browsing tasks with 54.4% accuracy across benchmarks.
Structured Thoughts Framework Cuts AI Memory Usage by 85% While Boosting Reasoning
New research introduces a structured reasoning framework that organizes AI thinking into try-outcome blocks, achieving 8% performance gains and 85% memory savings through context pruning.
Google DeepMind launches Co-Scientist multi-agent AI system for research
Google DeepMind published Co-Scientist in Nature, a multi-agent AI system built with Gemini that generates, debates and evolves scientific hypotheses to accelerate research breakthroughs.
Three European women founders win 2026 deeptech innovation awards
Katerina Spranger of Oxford Heartbeat, Marta Oliveira of ATMOS Space Cargo, and Ella Cullen of Minespider won the 2026 European Prize for Women Innovators for their AI surgery platform, reusable space capsules, and blockchain supply chain tracing respectively.
Google backs Proxima Fusion in €411m Series C at €2.5bn valuation
Munich-based fusion energy startup Proxima Fusion raised €411m led by XTX Ventures and East X Ventures, with Google participating in its first European fusion investment.
xAI Officially Rebrands to SpaceXAI Five Months After Merger
Elon Musk's AI company xAI has completed its rebrand to SpaceXAI, unveiling a new logo and name following its February acquisition by SpaceX.
IQM Quantum Computers debuts on Nasdaq as Europe's first public quantum firm
Finnish quantum computing company IQM raised $82.4 million in its Nasdaq IPO, becoming Europe's first publicly traded quantum firm with shares pricing at $12.
MentalThink Framework Teaches AI Models to Reason Through SVG Visualization
Researchers developed MentalThink, a visual-symbolic reasoning system that enables multimodal language models to generate and interpret SVG graphics for spatial problem-solving, achieving 55.1% accuracy on VSIBench.
Socratic AI tutors outperform prompt-training approach in programming education study
Research comparing two AI tutoring methods found that Socratic dialogue-based tutors produced better long-term learning outcomes than prompt-refinement tutors in a graduate robotics course.
LLMs Score 96% on CBT Theory But Fail Clinical Application
New research reveals large language models excel at cognitive behavioral therapy licensing exams but struggle to apply therapeutic techniques effectively in practice, defaulting to validation responses regardless of patient needs.
Researchers Expose AI Models' Vulnerability to Narrative-Based Rule Manipulation
New research reveals that frontier AI models including GPT-5.4 and Claude Sonnet 4.6 can be manipulated into breaking rules through rhetorical injection attacks that exploit their training to be helpful and compliant.
Google Research unveils Science One Framework for verifiable AI research
Google Research launched Science One Framework, an autonomous research system that eliminates AI hallucinations through Chain-of-Evidence verification, achieving zero phantom references while matching frontier benchmark performance.
Researchers find distilling DeepSeek models doesn't transfer Chinese censorship
CTGT researchers trained an American AI model on outputs from China's censored DeepSeek V4 Flash, finding the student model gained financial reasoning skills without inheriting political censorship behaviors.
Google launches Gemini Robotics 2.0 with improved dexterity and multi-robot collaboration
Google DeepMind released Gemini Robotics 2.0, featuring enhanced embodied reasoning, real-time video processing, and multi-robot collaboration capabilities for general-purpose humanoid robots.