Intel / article

Goodtriever Adaptive Toxicity Mitigation With Retrieval Augmented Models 2023 10 11

Source: cohere.com
Author: Cohere
Date: Published Jul 24, 2025

Context

Goodtriever steers a away from toxic text by consulting examples while it generates an answer. The research listed by Cohere addresses a problem with fixed safeguards: harmful language and its can change. The publisher abstract presents retrieval during generation as a way to make mitigation more adaptable.

The paper's method uses separate stores of toxic and non-toxic examples. For each next- decision, it retrieves similar contexts and adjusts the model's preferences toward non-toxic continuations and away from toxic ones. This makes the example collections part of the control mechanism. The researchers also test adaptation across different toxicity domains, making the work relevant to safeguards that must evolve with their deployment context.

Terms in this piece · Glossary

LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.

Read the source cohere.com