Sahaj.blogs

Day page

18 Feb 2026

Paper summary

Paper summary: GLM-5: from Vibe Coding to Agentic Engineering

GLM-5 targets "agentic engineering" beyond vibe coding, using DSA attention to cut long-context cost and an asynchronous RL infrastructure plus new agent RL algorithms for long-horizon tasks.

  • Generation is decoupled from training in its RL infrastructure, which the authors say drastically improves post-training efficiency.
  • The headline claim is real-world, end-to-end software engineering, not just code completion.

Read the post →

Study

LLMs · DPO

Direct Preference Optimization skips the separate reward model and RL loop and trains on preference pairs with a simple classification-style loss; the loss implemented and checked on a toy policy.

  • DPO turns preference learning into one supervised loss on log-probability ratios against a frozen reference model.
  • β controls how far the policy may move from the reference.

Read the post →