3 ms·
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 co
by smeej 4y ago
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.
- bordercases 4y agoHuh, how would that work?
- HN_is_for_gemes 4y agoHow would that work? Like you give the model a new DNA sequence and based on training it tells you if there is an error or not? Big question is why would the model do better than just regular search? Are the errors not regular?
- a_bonobo 4y agoStandard GPT models are not very good at comparing DNA-sequences, unfortunately https://www.biorxiv.org/content/10.1101/2023.03.11.532238v1 https://www.biorxiv.org/content/10.1101/2023.03.11.532238v1 See Figure 2, DNA alignment is mostly 0 accuracy