Published inTowards AIAlphaGeometry2: A Deep Dive into a Gold-Medalist AI Geometry SolverThe new model comes with interesting improvements in its architecture and pretraining techniques.2d ago2d ago
Published inTowards AIInside DeepSeek-R1: The Amazing Model that Matches GPT-o1 on Reasoning at a Fraction of the CostDeepSeek delivers another groundbreaking model.Jan 235Jan 235
Published inTowards AIInside rStar-Math, a Technique that Makes Small Models Math GPT-o1 in Math ReasoningThe new method represents an important evolution of reasoning for SLMs.Jan 13Jan 13
Published inTowards AIBuilding Large Action Models: Insights from MicrosoftA new framework for large action models published by Microsoft Research.Jan 71Jan 71
Published inTowards AIInside Deliberative Alignment: One of the Methods Poweing GPT-o3A method that teaches reasoning while following safety instructions.Dec 26, 2024Dec 26, 2024
Published inTowards AISome Insights About Phi-4: Microsoft’s New Small Foundation Model that Punches Above its WeightI recently started an AI-focused educational newsletter, that already has over 170,000 subscribers. TheSequence is a no-BS (meaning no…Dec 16, 20242Dec 16, 20242
Pretrain Your Own AI Models with Fast-LLMCreated by ServiceNow, the framework provides the key building blocks for pretraining AI models.Dec 12, 2024Dec 12, 2024
Published inTowards AIInside Anthropic’s Model Context Protocol (MCP)to Connect AI Assistants to DataThe protocol tries to standarize one of the most important elements of agentic applications.Dec 4, 2024Dec 4, 2024
Published inTowards AIInside Tülu 3: Allen AI’s New Post-Training FrameworkThe framework includes capabilities for data, models, post-training and evaluation under the same architecture.Nov 25, 2024Nov 25, 2024
Published inTowards AIInside FrontierMath: An Unprecedented Benchmark for Assessing Advanced Mathematical Reasoning in AIThe benchmark introduces evaluations that take AI mathematical reasoning to a new level.Nov 19, 20241Nov 19, 20241