New Terms, New Toxicity: Consensus-based Chinese Neologism Toxicity Detection via Search-Augmented LLMs
The episode discusses a paper on detecting toxic Chinese neologisms using consensus-based Search-Augmented LLMs. The research addresses how new or evolving slang bypasses moderation systems, showing that current commercial APIs miss over eighty percent of such content. The proposed SeTox framework uses search to provide real-time context, significantly improving detection rates from sixty-two percent to nearly ninety-three percent.