John Pearson
@jmxpearson.bsky.social
Computational neuroscience, neuroML, natural behavior. I charge more for miracles. PI @ pearsonlab.github.io. @dukemedschool.bsky.social.
Reposted by John Pearson
I just skimmed the paper (too fast...) and it seems like they're using more than the output sequence and actually the full logits of the output for reconstruction.
November 1, 2025 at 7:22 PM
I just skimmed the paper (too fast...) and it seems like they're using more than the output sequence and actually the full logits of the output for reconstruction.
Invertible in theory, but not at finite precision, no?
November 1, 2025 at 1:42 PM
Invertible in theory, but not at finite precision, no?