Thinking about this some more — I think content curation can work, we just need to think about it differently. So with ontolo's model, you have (random) content, and you ask users to label it. The assumption is that the content being presented is worth labeling (or even looking at) in the first place.

If we reverse the model, and ask people what they want to curate first, we can then use automation to narrow down content suggestions to then go into that bucket. This seems better because the curator has expressed interest in the subject matter, and has a much smaller pool of content to sift through. Instead of humans training AIs, AI could queue up higher value work for humans.

The interface could work something like this:

- What are you interested in? [categories/subcategories/custom label]
- What scope do you want to use? [follows|network|global|relay]
- Use heuristics, search terms, topics, and AI to present content to label. So for example, if they're curating pictures, don't show notes without a url in them.