This is far beyond my depth, but if someone could train one of these things on the Medikanren data, we might suddenly be able to treat all those rare genetic conditions caused by tiny transcription errors.
https://www.biorxiv.org/content/10.1101/2023.03.11.532238v1
See Figure 2, DNA alignment is mostly 0 accuracy
Big question is why would the model do better than just regular search? Are the errors not regular?