Classic ML to cope with Dumb LLM Judges

kept by maya sep 17

Combining many low‑confidence LLM relevance judgments via a decision tree boosts precision for e‑commerce search relevance, achieving up to ~96.5% precision on ~40% of test pairs.

OpenAI launches its first AI-native ad format
OpenAI introduces an AI-native ad format that lets users click to chat with a brand's AI agent inside ChatGPT instead of visiting the brand's website.
also on Wayfair
Under the Hood: Universal Commerce Protocol (UCP) - Google Developers Blog
Google and partners launched the Universal Commerce Protocol, an open-source standard for agentic commerce that standardizes secure, seamless shopping across AI surfaces.
also on Wayfair
2.7. Novelty and Outlier Detection — scikit-learn 1.9.1 documentation
The scikit-learn documentation explains how novelty and outlier detection differ, describing algorithms like Isolation Forest, Local Outlier Factor, and One‑Class SVM for identifying abnormal observations.
also on scikit-learn, #machine-learning, #developer-tools
Show HN: Timber – Ollama for classical ML models, 336x faster than Python
Timber compiles classical ML models into self-contained C99 artifacts, enabling 2 µs inference that is approximately 336× faster than Python.
also on scikit-learn, #machine-learning, #developer-tools