OpenAI Content Filter: the GPT-3 API moderation classifier
In one sentence During the GPT-3 API beta OpenAI provides a content filter, documented at least since October 2020, that classifies text as safe, sensitive or unsafe. It is not a dedicated endpoint but an engine (content-filter-alpha) queried through Completions; in 2022 the Moderation API replaces it.
During the GPT-3 private beta OpenAI offers developers a content filter: a model that reads a text, generated by GPT-3 or written by the user, and labels it safe, sensitive or unsafe.
The idea: before showing an output to a user, you run it through the filter, and if it is unsafe you block it or handle it otherwise. The API documentation has a page on the filter at least since October 2020.
The filter is free and, by OpenAI's own developers' admission, tuned to be cautious, with plenty of false positives. In August 2022 OpenAI replaces it with the Moderation API.
Companies
OpenAI
Tools
OpenAI Content Filter
Tags
Sources