Caltech Researchers Claim Compression of High-Fidelity AI Models

Posted by jonbaer on Wednesday April 01, 2026 @02:27AM
from the wsj.com dept.

https://www.wsj.com/cio-journal/caltech-researchers-claim-radical-compression-of-high-fidelity-ai-models-e66f31c9

( 3 comments | 2 points | on HN )

Re:Caltech Researchers Claim Compression of High-Fidelity AI... by WalterGR on Wednesday April 01, 2026 @03:59AM (#47598128)

105 comments: https://news.ycombinator.com/item?id=47593422

“Show HN: 1-Bit Bonsai, the First Commercially Viable 1-Bit LLMs” (prismml.com) | 227 points | 10 hours ago

[ Parent | on HN ]

Re:Caltech Researchers Claim Compression of High-Fidelity AI... by grimm8080 on Wednesday April 01, 2026 @02:30AM (#47597566)

I need a subscription to read that. Though, I don't get why 1 bit LLMs are smaller, if you go from 8 bit to 1 bit, you'll need 8 times more nodes right?

[ Parent | on HN ]

Re:Caltech Researchers Claim Compression of High-Fidelity AI... by jonbaer on Wednesday April 01, 2026 @02:35AM (#47597603)

https://archive.ph/XIYG5

[ Parent | on HN ]