Classifier
A small model with one job: sort something into categories, like safe or dangerous.
A classifier is a model that does not write anything. It looks at an input and picks a label from a short list. Is this email spam or not. Is this image a cat or a dog. Is this command about to delete something important. That narrow job means a classifier can be much smaller and cheaper than the model doing the main work, so it can run constantly in the background without slowing things down.
Classifiers have become the standard safety layer around AI systems. A big model proposes an action, a classifier checks it, and only the flagged cases reach a human. It is a useful design because a classifier never gets tired or bored, which is exactly where human reviewers fail. It is not a guarantee though. A classifier that catches 89 percent of problems still misses one in nine, so the sensible setup treats it as a filter, not a fence.
-
Watching Its Own Model Now Costs OpenAI 20 Percent Extra, and It Paused a Training Run to Do It
-
Claude Code Turns On Auto Mode by Default, and the Safety Numbers Are Not What You Would Guess
-
Anthropic loosens Claude Fable 5's biology filter, cutting blocked questions by 85 percent
-
Substack added an AI detector, and its writers are not having it
-
Anthropic's Opus 5 is smaller and cheaper, yet it beats its bigger sibling
-
Claude Code Gets a Built-In Browser That Can Read, Click, and Type for You
-
Using Fable 5 with a Claude subscription? Monday changes how you pay