AI model - false hit

jello234

Like everyone else, I'm just thrilled with this new capability. Of course, there are growing pains, but they are mostly bearable. One thing I noticed is that by including indexes, it may not be the best use of the AI model. For example, I ran a query on "gideon walch." I got one hit:

https://www.familysearch.org/search/full-text/results?count=100&q.text=%22gideon%20walch%22

When I looked at the hit, which pointed to the manual index for that roll of microfilm, it looked to me like "hatch" vice walch. And, sure enough, as I dug into that roll and pinpointed the actual reference, it was "hatch."

It appears that the model is doing a pretty decent job when there is a significant amount of a particular writer's handwriting - which could be why it didn't see "walch" in the body of that microfilm. But, it doesn't do so well when there may be different handwriting involved on a page.

This brings me back to previously reported problems I've experienced where I'm getting a lot of results for indexes, but not the main search term I'm looking for. So, I wonder, might it be possible to either eliminate indexes (for now) from the AI search model? Of course, that wouldn't work against a collection where a manual index is part of the body of records. But, it might be helpful to exclude collections which are simply indexes to eliminate a higher probability of false hits. Again, as a "for now" solution. Alternatively, perhaps create a toggle for users to exclude indexes themselves.

p.s. forgot to mention I'm using FF browser (122.0, yes, I know I need to update to 124.0).

RobertWessel

@jello234

Thank you for your comments. We are pleased with the response that we have received on this search tool. However, we also recognize that it is in its infancy and fully expect it to improve over time as it continues to l earn and as we add new features.

Keep youur eyes open for announcements of new features, which will be announced on the landing page for the tool as well as in this forum.