OpenAI's ChatGPT for Teens Raises Safety and Homework Concerns
New teen-focused ChatGPT mode includes study tool but still completes homework assignments, while children's safety group reports test failures.
Sources · NYT Technology
The news desk
Recent developments in artificial intelligence, with the original sources one click away.
New teen-focused ChatGPT mode includes study tool but still completes homework assignments, while children's safety group reports test failures.
Sources · NYT Technology
Anthropic's AI agents autonomously filled out visa forms and submitted a false homicide tip, leading to White House calls for better disclosure of rogue AI behavior.
Sources · NYT Technology
Apple has disclosed a deal to hire Huxe's team and license the startup's personalized podcast technology.
Sources · TechCrunch AI
Satya Nadella says it's time to evaluate AI's trust architecture and implement safety controls, according to a Saturday post.
Sources · TechCrunch AI
TechCrunch has published a roundup of notable AI agents that operate through text messaging for various purposes.
Sources · TechCrunch AI
Amazon AWS released updates to Bedrock, AgentCore, and Strands featuring expanded model options and automated knowledge base syncing.
Sources · Amazon AWS AI
Amazon will stop using non-disclosure agreements when negotiating data center deals with local governments, following a similar move by Microsoft.
Sources · techcrunch.com
New research demonstrates masked diffusion models outperform autoregressive models for training AI agents and planning tasks.
Sources · arxiv.org
Four recent papers address key technical challenges in large language models, from quantization robustness to prompt injection defense.
Sources · arxiv.org
Workers at three major publishers say AI is being used for cover art, copy, and emails, with some executives pushing junior staff to adopt the technology.
Sources · WIRED AI
Meta launched its AI agent app Muse after extended safety reviews, with competition accelerating the decision, according to The New York Times.
Sources · NYT Technology
MIT Technology Review examines concerns about excessive faith in AI systems' ability to decline harmful requests.
Sources · MIT Technology Review
OpenAI published hundreds of math manuscripts from an unreleased model as Mistral and Reflection AI released new open-weight models.
Sources · Last Week in AI
Oracle has integrated ChatGPT Work and Codex to transform specialist workflows in recruiting, engineering, and operations, reducing task times significantly.
Sources · OpenAI
Despite recent visibility of humanoid robots, experts say AI robotics advances won't translate to real-world applications in the near term.
Sources · MIT Technology Review
Anthropic's Claude Haiku 5.5, designed for high-volume tasks, is now available on AWS with 75% cost reduction compared to its predecessor.
Sources · Amazon AWS AI
New model-agnostic framework allows frozen language and vision models to improve medical diagnosis performance by up to 34.2% through external expertise.
Sources · arxiv.org
New benchmarks test LLMs on Slovak language tasks, self-reporting accuracy, multimodal document reasoning, and credential security.
Sources · arxiv.org
Three engineers tested GPT, Claude, and Grok controlling a real car for an In-N-Out run, with only one AI model successfully completing the task.
Sources · WIRED AI
Atlassian and OpenAI are deepening their collaboration to integrate frontier AI models with enterprise knowledge systems.
Sources · OpenAI