Meta's Superintelligence Lab released Muse Wonder, its first frontier model, with a heavy emphasis on long-term memory.
AI Keeps Acing Math Tests. MIT Just Built a Harder One — 30,000 Olympiad Problems From 47 Countries.
MIT CSAIL, KAUST and HUMAIN built MathNet — 30,000+ expert-authored Olympiad math problems from 47 countries and 143 competitions, a…
MLCommons escalates AI benchmark with Meta’s colossal Llama 3.1 model
MLCommons updates MLPerf Training benchmark to Meta’s 403B-parameter Llama 3.1, quadrupling dataset requirements amid industry race for larger AI models.…
Benchmark Wars: Stanford Reveals AI Model Gap Narrows to Statistical Noise
Stanford’s 2024 report shows closed/open AI model performance gap collapsed from 15.9% to 0.1%, challenging traditional evaluation methods and reshaping…
OpenAI Launches GPT-4.1 Family With 83% Cost Cut, Targets Developer Ecosystem
OpenAI unveils GPT-4.1 model family featuring API-exclusive access, 1M-token context windows, and enhanced coding capabilities, intensifying competition with open-source AI…
Meta’s Llama 4 AI model sparks US-China tech arms race: Can open source secure America’s lead?
Meta’s release of Llama 4, a 17B-parameter open-source AI model outperforming GPT-4o, reignites debate about US AI leadership amid China’s…
Nvidia dominates MLPerf AI benchmarks with Blackwell architecture
Nvidia’s Blackwell GPUs lead in MLCommons’ latest AI benchmarks, outperforming AMD and highlighting the importance of inference performance for enterprise…
What does ‘PhD-level’ AI mean? OpenAI’s $20,000 agent plan explained
OpenAI’s rumored $20,000-per-month AI agent, marketed as ‘PhD-level,’ raises questions about its value, reliability, and real-world applicability compared to human…