Tag
Microsoft introduces MAI-Cyber-1-Flash, a compact security model for MDASH, achieving 96% on CyberGym with 50% cost savings compared to previous models. The model is designed to handle 90% of security tasks efficiently, enabling use of larger models for harder tasks.
The article criticizes Microsoft AI's lackluster coding model performance compared to rivals like Kimi K3 and Deepseek V4, suggesting MAI is far behind despite vast resources.
Ed Zitron critiques Microsoft AI as decrepit, disconnected, and directionless, predicting a pending implosion for the tech giant's AI initiatives.
The article explains how GEPA (Genetic-Pareto Optimization) within DSPy is used for efficient prompt tuning, specifically applied to pretraining data curation at Microsoft AI, allowing researchers to replace manual prompt engineering with automated compute-driven optimization.
Microsoft AI CEO Mustafa Suleyman discusses the near-term possibility of superintelligence, the company's restructured relationship with OpenAI, and new frontier models, asserting that AI will not replace human jobs.
The tweet discusses Microsoft AI's use of Ray actors for training the MAI-Thinking-1 model, enabling finer granularity for heterogeneous compute and better CPU resource utilization in GPU clusters.
GEPA-optimized LLM judges from dspy are used for data filtering in Microsoft's MAI-Thinking-1 model pre-training pipeline.
Microsoft AI introduces MAI-Thinking-1, a 35B-active parameter reasoning model trained from scratch without distillation, achieving strong performance on software engineering and math benchmarks while emphasizing clean data and self-sufficiency.
Mustafa Suleyman, CEO of Microsoft AI, predicts that within 18 months, AI will automate most white-collar tasks involving computer work, including accounting, legal, and project management. The article discusses contrasting views on AI's actual impact on professional jobs.