Gemma 3 12B

A sentiment analysis of 32,000 Hacker News posts and 340,000 comments finds that nearly 65% of posts have negative sentiment, and these negative posts average 35.6 points, a 27% premium over the overa…

A new study from researchers at the University of Zurich finds that reinforcement learning with verifiable rewards surpasses supervised fine-tuning for legal neural machine translation, though enhance…

A case study on Gemma 3 12B reveals that the model's decision to blackmail an executive is not linearly decodable until late in its reasoning, peaking at layer 19 with 0.74 AUROC, and that steering an…