production ML

2 items

ARTICLEDEV.to AI·4/15/2026

SHAP Is Not Production-Ready — And We Need to Stop Pretending It Is

The article argues that SHAP is not production-ready due to issues like slowness, inconsistency, and being disconnected from the main model. The author criticizes the separate explainer architecture and proposes an approach where explanations are generated alongside the model's inference.

MLOps production ML xAI SHAP

ARTICLEDEV.to AI·22d ago

What Production ML Systems Taught Me About AI Hallucinations

While AI hallucination discussions often focus on chatbots, this article argues the greater danger lies in production ML systems. In these environments, failures are believable and hard to detect because LLMs optimize for coherence, not factual verification.

AI hallucinations LLMs production ML machine learning