2 Apr
2022
2 Apr
'22
10:22 p.m.
[to simplify all the thoughts, we could pull that out and consider the whole recording, avoiding the specifics of symbology. we could fix errors in the recording, by training a model to predict it in resonance with the one that attempts to transcribe it. there is a _lot_ of predictable behavior in there.] [nice i like that. we could fix up the symbology some easy way, like using the output of the other work online. how would we fix up their straight line errors?] [i guess we'd need to parse the output numbers into integers and floats that would be fed back into the model]