FLUX3D paper presents method for generating high-fidelity 3D content using diffusion-aligned sparse representations, advancing 3D generative AI.
Leaked files reveal systematic Russian efforts to contaminate AI training datasets and search indices through coordinated fake platform creation.
New benchmark DeepSWE specifically measures how well frontier AI models perform at actual software engineering tasks. Provides empirical evaluation beyond generic coding benchmarks.
Mozilla used Anthropic's AI model to identify and remediate 271 bugs in Firefox, demonstrating production-scale security application of AI code analysis.