TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Deciphering language processing in the human brain through LLM representations

2 pointsby simonpureabout 2 months ago

1 comment

cadamsdotcomabout 2 months ago
“The brain has areas that encode and decode speech”<p>“They activate before speaking and comprehension at different, but predictable, times”<p>“We used deep learning but no one currently cares about DL research so we manufactured a link to LLMs”<p>The problem with brain research using electrodes is patterns of neural activation is too blunt an instrument to map to the specifics of what people want to learn is going on - mostly thanks to the thickness and effectiveness of the blood-brain barrier as an EM shield - and that doesn’t seem to change even with DL, or we’d have seen more breakthroughs. And brain implant tech is still a ways away. And even an implant would only give detailed info about the local area it’s implanted, and the implantation process may damage key bits of tissue and risks altering the region of the brain under test.<p>This type of research, with electrodes strapped to a person’s head, is kind of like standing in the next room trying to guess the song you can hear through the wall.