Using "summarize" to populate docs page summaries
In this November 2023 Slack thread I asked about using summarization features to populate metadata in our documentation files. The discussion follows:
Amy: Our docs pages lack summaries that can be used to populate preview cards in Slack, Signal, etc. For example, https://docs.gitlab.com/ee/user/project/merge_requests/ai_in_merge_requests.html will show the default description text (GitLab product documentation) unless we add a description: field to the page's metadata. While I can generate that by hand, what if I could feed the page contents to our local friendly bot and get a 1-3 sentence summary in return? Even a partially-automated solution could save significant time given how huge our docs site is:
- paste in page URL
- get summary
- edit, and manually transfer to page
- make merge request
- profit
@tmccaslin: This would be a new feature request. But you could accomplish this with the claude console today, just pasting in the markdown and asking it for a sentence summary produces this:
Taylor: You can do an access request to get access to claude. While we haven't done this widely, I think this could solve your problem today and help us generate some use cases for TW productivity ideas. Anthropic Claude is approved for use by legal to contribute code back to gitlab.
Amy: Let me pass this info over to my manager. I don't want to burn too much of my time without sharing it with her, but that's a meaningful step forward.
@phikai: I would think some kind of rake task to generate them would be useful...then you could do it in bulk or by specific file, send it for an MR and just edit as needed.
Taylor: realistically we're probably to a point where with the AI framework stable, we could introduce LLM powered rake tasks. Really just need a someone who can do some rake ruby coding. The hardest part is probably just managing the API call to the AI framework and ensuring it's secure.
(Capturing this information in an issue because Slack messages are ephemeral. I've marked it confidential for now because I'm not sure how much of this info is SAFE.)