sinanuozdemir/oreilly-agi

It provides practical guidance for integrating and extending AGI models, demonstrating how to add short-term memory to OpenAI agents and enable reasoning models to control a laptop. The project includes extensive benchmarking notebooks that run various AGI models against datasets like Humanity's Last Exam and MMLU, alongside case studies on applying reinforcement learning (GRPO) with models like Qwen. It also covers advanced prompt engineering techniques for DeepSeek-R1, leveraging few-shot learning and vector databases.