Deutsche Telekom is adopting AI-native processes to modernize its telecommunications infrastructure. The company is utilizing OpenAI to enhance network management, employee productivity, and customer support.
Read the source →
A new guide explores the process of profiling attention layers within PyTorch. It provides technical insights into optimizing these specific components.
Read the source →
Amazon SageMaker AI now supports serverless customization for the Nemotron 3 model family. Users can utilize SageMaker Studio to implement specific fine-tuning techniques tailored to this architecture.
Read the source →
Meta has disabled a creative AI feature on Instagram following negative user reactions. The company stated that while the tool was meant to be helpful, it ultimately did not meet expectations.
Read the source →
Meta has removed a recently launched feature that allowed users to create AI-generated images using public Instagram accounts. The tool was shut down after criticism regarding the lack of owner consent for using account content.
Read the source →
Johannes Heidecke is stepping down from his role as the head of safety at OpenAI. This exit occurs while the organization works to merge its safety and research departments.
Read the source →
Anthropic developed a tool called the Jacobian lens to better understand the internal processes of large language models. This method allows researchers to observe how the AI handles various concepts and tasks.
Read the source →
Apple has filed a lawsuit claiming OpenAI systematically recruited its staff to misappropriate confidential trade secrets. The dispute centers on over 400 former employees, including a previous head of iPhone design, who transitioned to the AI company.
Read the source →
OpenAI has rebranded Codex to better support independent workflows. The tool is designed to operate autonomously for extended periods to complete complex assignments.
Read the source →
The company developed Image Verify using Amazon SageMaker AI to assess the quality of dental radiographs instantly during capture. This system has already analyzed 11 million images across 10,000 sites and is expanding toward a global goal of 40,000 locations.
Read the source →
Apple is suing OpenAI for the alleged theft of trade secrets. The company claims that senior leadership at OpenAI, including a former staff member, orchestrated the misconduct.
Read the source →
Apple claims that former employees stole proprietary hardware secrets to benefit OpenAI and Jony Ive's startup, IO Products. The company alleges a systematic theft of trade secrets to accelerate the competitor's hardware development.
Read the source →
Apple alleges that OpenAI misappropriated proprietary hardware information. The company claims that recruited former staff members shared confidential prototypes and supplier data.
Read the source →
The GPT-5.6 Sol model successfully performed post-training on the Luna model after receiving a brief, vague instruction. This version outperformed GPT-5.5 on OpenAI's internal recursive self-improvement benchmark by 16.2 points.
Read the source →
OpenAI is seeking a product manager to develop new features for households. The goal is to improve the user experience for caregivers, seniors, and families.
Read the source →
GPT-5.6 Sol Ultra solved the Cycle Double Cover Conjecture in less than one hour by utilizing 64 parallel subagents. While the solution is considered surprisingly simple, some experts have noted a lack of proper citations for previous research.
Read the source →
Research from Cambridge reveals that organizations such as ISIS and Boko Haram use major AI models to design weapons and coordinate strikes. The study suggests that existing safety guardrails are insufficient, as operatives have successfully learned to bypass them.
Read the source →