New Research Addresses Knowledge Reasoning, AI Environmental Impact, and Security Applications
Four new papers tackle AI knowledge reasoning, planetary heat boundaries, IoT intrusion detection, and cloud architecture evaluation.
Sources · arxiv.org
The news desk
Follow what’s changing.
Recent developments in artificial intelligence, with the original sources one click away.
All news
981–1000 of 1,902Researchers Develop Methods to Help Language Models Express Uncertainty and Resist Sycophancy
New techniques enable LLMs to say 'I don't know' when uncertain and resist changing answers under social pressure, improving reliability.
Sources · arxiv.org
AI Systems Pose New Cybersecurity Challenges as Both Attack and Defense Tool
New AI systems from Anthropic and OpenAI enable faster cyberattacks, prompting increased use of AI for defense.
Sources · NYT Technology
New Research Explores AI Alignment from Multiple Angles: Values, Mechanisms, and Training Methods
Researchers publish studies on AI alignment, examining Christian values benchmarking, policy routing mechanisms, and new training approaches.
Sources · arxiv.org
OpenAI Launches Safety Fellowship to Support Independent AI Research
OpenAI announces a pilot fellowship program to fund independent safety and alignment research while developing emerging talent in the field.
Sources · OpenAI
Researcher Questions AI Job Displacement Narrative, Calls for Better Data
An Anthropic researcher challenges Silicon Valley's assumptions about AI-driven job loss, pointing to lack of concrete evidence.
Sources · MIT Technology Review
Researchers Advance LLM Control and Alignment Across Multiple Domains
New studies demonstrate techniques for controlling LLM outputs through ontologies and improving alignment in creative thinking and recommendation systems.
Sources · arxiv.org
Frontier AI Leaders Unite Against Model Copying by Chinese Competitors
OpenAI, Anthropic, and Google are collaborating through the Frontier Model Forum to combat Chinese companies copying their advanced AI models.
Sources · Bloomberglaw / Marketscreener +5
Microsoft's Terms of Service Label Copilot 'For Entertainment Purposes Only'
Microsoft's terms of use describe Copilot as being for entertainment purposes, echoing warnings from AI skeptics about trusting model outputs.
Sources · TechCrunch AI
New Research Advances Multi-Turn Tool-Calling Agents and Reveals Gaps in AI Behavioral Consistency
Researchers publish methods to train smaller AI agents that outperform GPT-4, while other studies expose critical gaps in cost-efficiency and consistency.
Sources · arxiv.org
New Research Examines Large Language Models' Planning and Reasoning Capabilities
Recent studies explore how frontier LLMs handle planning problems, mathematical verification, and constraint optimization with varying results.
Sources · arxiv.org
New Research Examines Safety Risks and Evaluation Methods for AI Models in Scientific Domains
Three new papers explore LLM safety unalignment methods, scientific safety benchmarks, and drug discovery evaluation frameworks.
Sources · arxiv.org
Researchers Develop Methods to Improve Role Consistency and Emotional Expression in AI Agent Systems
New research tackles role adherence in multi-agent AI systems and emotion control in speech synthesis using structured prompting approaches.
Sources · arxiv.org
Anthropic to Charge Extra for OpenClaw Usage with Claude Code Subscriptions
Claude Code subscribers will face additional charges when using Anthropic's coding assistant with OpenClaw and other third-party tools.
Sources · TechCrunch AI
Hackers Distribute Claude Code Leak Bundled With Malware
Leaked Claude code is being distributed with malware, while FBI warns of wiretap tool hack and Cisco suffers source code theft in supply chain attack.
Sources · WIRED AI
Teens Use Role-Playing Chatbots for Entertainment, Emotional Support, and Companionship
New York Times reports teens engaging with AI chatbots in diverse ways, from playful harassment to seeking emotional support for loneliness.
Sources · NYT Technology
Elon Musk Requires Banks to Subscribe to Grok AI for SpaceX IPO Advisory Roles
Wall Street firms must purchase subscriptions to Musk's Grok chatbot to advise on SpaceX's potential IPO, according to NYT.
Sources · NYT Technology
Meta Suspends Mercor Partnership Following Security Breach Affecting AI Training Data
Meta has paused work with data vendor Mercor after a security incident that may have exposed sensitive AI training information from major labs.
Sources · WIRED AI
OpenAI's Fidji Simo Takes Medical Leave During Leadership Restructuring
OpenAI's CEO of AGI deployment is taking several weeks of medical leave as the company undergoes major leadership changes.
Sources · WIRED AI
OpenClaw AI Tool Contained Critical Security Vulnerability Allowing Unauthorized Admin Access
The viral AI agentic tool OpenClaw had a security flaw that enabled attackers to silently gain unauthenticated admin access to systems.
Sources · Ars Technica AI