Thomas Vander Wal<p><span class="h-card" translate="no"><a href="https://mastodon.social/@x28de" class="u-url mention" rel="nofollow noopener" target="_blank">@<span>x28de</span></a></span> <span class="h-card" translate="no"><a href="https://mastodon.social/@jarango" class="u-url mention" rel="nofollow noopener" target="_blank">@<span>jarango</span></a></span> This is how I've used <a href="https://mastodon.social/tags/devonthink" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>devonthink</span></a> for over 10 years (for this and used it over 20) and it works quite well to bring back things I've read of (mostly likely) gutted. I have between 70k and 100k objects in it and its relevance confidence has always been incredibly good. </p><p>I'll use LLMs to try and point to relative elements. But, context and linguistic shadings, as well as cross domain relevance detection is insanely poor and those are areas I wade in often.</p>