Overview: Stanford CME295 Transformers and Large Language Models
Stanford's CME295 course on Transformers and Large Language Models will cover critical advancements in agent systems, harness engineering, diffusion LLMs, and on-policy distillation. These topics represent significant progress made over the past year and form the core curriculum for the upcoming offering. The course aims to equip professionals with the latest technical frameworks and methodologies driving the field forward. Enrollment is open for those seeking to master these evolving technologies.
DeepSeek Just Open Sourced Their Own Claude Code!
DeepSeek has released the DeepSeek Harness, an open-source agent framework licensed under MIT that utilizes a modular Cordis architecture where every component functions as a plugin. This system allows developers to customize models, tools, sandboxes, and UIs while providing full traceability of agent actions through a dedicated trajectory section. The platform supports multiple runtime modes including standard, code, minimal, and creator modes, enabling flexible deployment for both basic and advanced development tasks. Although currently in preview, the tool offers immediate access to various model providers and requires only Node.js and an API key for local setup.