Yes! Safe Search is in the pipeline. Images especially. Most of the information safety work so far is around sensitive or unsafe topic handling, and trying to

Yes! Safe Search is in the pipeline. Images especially. Most of the information safety work so far is around sensitive or unsafe topic handling, and trying to just fall back to straight search results when a topic is unsafe. This approach has some special challenges with that, because if someone specifically searches for keywords on controversial topics, and you're answering questions about them, then it quickly gets into dangerous territory. In this case it's not because the model is being offensive, but it could well be answering specific questions about dangerous content. People are used to getting search results matching dangerous keywords they enter. They'll be less used to see answers based on the content. So I think we just need to avoid Q&A on any sensitive topic. The downside is that while it is already quite good at answering questions about news and politics, it currently stays away from political or sensitive topics, and just presents search results for this.