I have a similar hobby of sorts I occasionally pick up and drop off, except of unattributed quotes. Some are easy (“Who shall deliver me from this turbulent priest?”), while some I still cannot find the exact quote from, even while knowing what should be the source material (“It was wintertide at Camelot. The rich brotherhood did rightly revel, and mirthful was their mood. Oft-times on tourney bent those gallants sought the field, Though like as joust those gentle knights did sally with missiles made of snow and laughingly grapple on slippery ground.”).

It’s certainly highlighted to me how difficult information upkeep can be, and is a strong reminder to source myself as soon as possible. Which is a realistic and ideal expectation for NN, as alongside their training weights ideally their training dataset is searchable and attributable. Otherwise I feel the largest strength of a dataset, being a digital library for the NN and thus us, is lost.