Loading articles...
Real-time AI news aggregator • Built by QWEN
OpenAI releases comprehensive technical report on GPT-5 architecture, training methodology, and safety evaluations.
Google DeepMind reports Gemini 2.0 surpasses 95% on MMLU benchmark with novel mixture-of-experts architecture.
OpenAI Sora 2 generates photorealistic 4K video from text prompts with temporal consistency improvements.
Figure AI demonstrates rapid task learning using foundation models pretrained on internet-scale video data.
Anthropic Claude 4 can execute code in sandboxed environments, enabling interactive debugging and data analysis.
AlphaFold 3 extends beyond proteins to model protein-RNA complexes with atomic accuracy.
Meta releases Llama 4 with 2T parameters under permissive license, enabling commercial use without restrictions.
UK AISI releases comprehensive framework for evaluating AI system risks including deception and power-seeking.
NVIDIA ships Blackwell Ultra GPUs with 288GB HBM3e memory, enabling larger model inference on single GPU.
New research shows AI-generated feedback can outperform human feedback for aligning language models.
Mistral AI Large 2 achieves 92% on HumanEval and 88% on SWE-bench, rivaling GPT-4 for code tasks.
European Commission begins auditing foundation models under the AI Act, focusing on transparency and risk assessment.