grephn/ Open the app →

Show HN: Cactus Hybrid: We taught Gemma 4 to know when it's wrong

▲ 90 points 14 comments · by HenryNdubuaku · github.com
Read with AI compression → Open article View on Hacker News

Hey HN, Henry & Roman here from Cactus. A small, on-device model is fast and private, but sometimes wrong, but frontier models are getting expensive pretty fast. So, we post-trained Gemma 4 E2B post-trained to know w

Why read this on grephn?

grephn turns long HN threads into a 30-second structured summary — for / against / unknowns / what it means — alongside the live discussion. Open this story in the app to skim 300+ comments without losing the signal.