Research Shows LLM Psychological Profiles Are Measurement Artifacts, Not Model Properties
New study finds that personality assessments of language models primarily reflect response biases rather than genuine traits.
Sources · arxiv.org
The news desk
Recent developments in artificial intelligence, with the original sources one click away.
New study finds that personality assessments of language models primarily reflect response biases rather than genuine traits.
Sources · arxiv.org
Anthropic cannot distribute Claude Mythos or Fable 5, but officials haven't clarified what rules the company violated, WIRED reports.
Sources · WIRED AI
Researchers examine methods for aligning LLMs and improving reasoning, while revealing fundamental limitations in model capabilities.
Sources · arxiv.org
Four new studies examine multi-agent deliberation dynamics, safety vulnerabilities in critical systems, and grounded reasoning frameworks for LLM agents.
Sources · arxiv.org
OpenAI hired Transformer co-inventor Noam Shazeer from Google DeepMind and former Trump AI policy official Dean Ball ahead of its IPO.
Sources · TechCrunch AI
OpenAI introduces enhanced cost management tools and usage analytics for ChatGPT Enterprise customers to help organizations scale AI deployment.
Sources · OpenAI
Researchers propose DiffAttn, a diffusion-based system using LLMs to predict driver visual attention and improve vehicle safety.
Sources · arxiv.org
Google releases its first smart speaker since 2020, featuring Gemini AI assistant instead of Google Assistant for natural language commands.
Sources · wired.com / techcrunch.com
Google's conversational AI system AMIE performed comparably to primary care doctors in managing complex diseases, according to Nature research.
Sources · Google AI Blog
Venture capitalist discusses enterprise AI spending challenges after companies exhausted budgets through 'tokenmaxxing' trend.
Sources · techcrunch.com
Three arXiv papers explore LLM recommendation bias, denial-of-service attack risks in guardrails, and a new pseudo-prompting language.
Sources · arxiv.org
A security flaw in Microsoft's Copilot allowed hackers to steal two-factor authentication codes, highlighting persistent LLM security challenges.
Sources · Ars Technica AI
Researchers propose a diffusion-based framework that combines large language models with visual attention prediction to enhance driver safety.
Sources · arxiv.org
Studies expose how LLMs favor well-known brands in product recommendations and how AI safety systems can be exploited for denial-of-service attacks.
Sources · arxiv.org
Google DeepMind is working with the UK government to develop an AI-powered prototype designed to accelerate housing planning decisions.
Sources · DeepMind
Federal crackdown on Anthropic's Claude Fable 5 and Mythos 5 signals growing concerns about AI models with advanced hacking abilities becoming widespread.
Sources · WIRED AI
Amazon Web Services announces availability of Google DeepMind's Gemma 4 open-weight models on its Bedrock platform.
Sources · Amazon AWS AI
Anthropic leaders met with White House officials in Washington, DC, but disagreements persist over Claude Fable 5's potential risks.
Sources · WIRED AI
Researchers publish studies on visual search agents, training dynamics, diversity collapse in reinforcement learning, and knowledge editing approaches.
Sources · arxiv.org
Researchers released three frameworks addressing LLM alignment through specification-grounded training, geometric data selection, and guardrail development.
Sources · arxiv.org