This is an interesting article by Niall Murphy (O'Reilly author, SRE legend). However, I think has a few unsubstantiated claims for an academic article (e.g., "it's quite hard to have meaningful alerting for ML systems" - I work in this space, know this isn't true in practice).
It is also very agent/LLM-centric, without differentiating between LLMOps and MLOps. For example, the whole area of evals for agents is conspicuous by its absent.
评论
1 条预览评论 · 正在加载完整讨论请先登录 h4cker 账号，然后连接 Hacker News 后发表评论。
This is an interesting article by Niall Murphy (O'Reilly author, SRE legend). However, I think has a few unsubstantiated claims for an academic article (e.g., "it's quite hard to have meaningful alerting for ML systems" - I work in this space, know this isn't true in practice). It is also very agent/LLM-centric, without differentiating between LLMOps and MLOps. For example, the whole area of evals for agents is conspicuous by its absent.