OpenAI Finds 30 Percent of Popular AI Coding Test Broken
OpenAI has pulled its endorsement of SWE-Bench Pro after finding 30 percent of its tasks flawed, impacting AI performance assessments.
Archive
Browse all published articles.
ARCHIVE
-
-
Anthropic, OpenAI, and SpaceX Set to Outperform 25 Years of Tech Exits
The pending IPOs of Anthropic, OpenAI, and SpaceX are projected to generate more value than all U.S. VC-backed exits since 2000, according to a recent report.
-
IBM Introduces CofrGenets to Improve Transformer Model Training
IBM researchers have developed CofrGenets, a new framework that reduces training errors in transformer-based models by 40%.
-
AMD Analyzes AI Network Traffic Patterns and Congestion
AMD details AI training network traffic trends, showing congestion peaks and harmonic-like patterns in large-scale clusters.
-
AMD Introduces FlyDSL for High-Performance GPU Kernel Development
AMD released FlyDSL, a Python DSL for GPU kernel development, enabling performance matching hand-tuned C++ kernels with reduced complexity. The tool is tested on AMD Instinct MI355X GPU using ROCm 7.2.2.
-
Mistral Studio Offers Governance for AI Prompts and Skills
Mistral announced on July 9, 2026, that Studio provides a centralized system of record for managing AI prompts and skills, ensuring traceability and compliance.
-
Meta Launches Muse Image and Muse Video Models
Meta introduced Muse Image and Muse Video, its first media generation models, with Muse Image ranked second in text-to-image benchmarks as of July 5, 2026.
-
Anthropic Introduces Claude Reflection Feature
Anthropic launched a new reflection tool for Claude users, offering insights into AI integration in daily life, available in beta for Free, Pro, and Max users.
-
Databricks Uses GLM 5.2 as Default Coding Engine
Databricks found GLM 5.2 performs on par with Opus 4.8 but at a lower cost, leading to its adoption as the default coding model.
-
Nandan Nilekani Steps Down as Fundamentum Launches Third $200M Fund
Nandan Nilekani, co-founder of Infosys, is stepping down from his general partner role at Fundamentum as the firm launches its third $200 million fund.
-
Ollama Raises $65M in Series B, Grows to Nearly 9M Users
Ollama, the open-source AI developer tool, raised $65 million in Series B funding, growing its user base to nearly 9 million.
-
OpenAI AI Dominates AtCoder Contest, Solving All Five Problems
OpenAI's AI system outperformed all human competitors at the 2026 AtCoder World Tour Finals, solving all five Algorithm Division problems, including two rated exceptionally difficult.