This labelling approach — when I use a model, say OpenAI’s, or Claude’s, or Gemini’s, there’s a thumbs-up or thumbs-down under the conversation. What’s different? Why do we need this?