GTFormer: What If Long Context Didn’t Need Attention At All...
A new preprint trains a model on just 512 tokens and runs it on 4 million without any fine tuning. It does this by dropping self attention completely.
D
2 min read
Issues
CSI Tech Weekly: Week of Oct 30, 2026
Share a technical deep-dive, system design breakdown, or engineering tutorial with the CSI MIT-WPU community.
When a Certificate Authority’s Keys Turned Out to Be Guessable
Chrome, Firefox, Apple and Microsoft all cut off certificates from Bengaluru-based eMudhra within days. The cause was a set of keys that should have been random and weren’t.
L
3 min read