How confessions can keep language models honest

Company news OpenAI Source language: English

OpenAI researchers are testing “confessions,” a method that trains models to admit when they make mistakes or act undesirably, helping improve AI honesty, transparency, and trust in model outputs.

Excerpt

OpenAI researchers are testing “confessions,” a method that trains models to admit when they make mistakes or act undesirably, helping improve AI honesty, transparency, and trust in model outputs.

Short excerpt from the source — open the original for the full text.

Sources

← Back to the daily brief