Hacker News Reader

Yes friend but you see the openai doesn't care they don't have enough labour to filter the bad apples. If the world heads towards destruction, it has been because we were mean to chatgpt and it trained him further.

View on HN

I don't think it would require manual labor. AI research labs like OpenAI, Anthropic, Alphabet's Gemini team, etc already make extensive use of LLM's internally, and there has been quite a bit of work already done on models to detect toxicity in text, which could simply be inserted between the message router and the actual inference initialization on the user's prompt at very little computational cost.

See: Google's Perspective API, OpenAI Moderation API, Meta's Llama Guard Series, Azure AI Content Safety API, etc.

Reply on news.ycombinator.com