AI Briefing#gemma-3
The latest AI and developer news about #gemma-3, with the original source and a short summary.
The latest AI and developer news about #gemma-3, with the original source and a short summary.
A study on Gemma 3 270M shows that training just 1,000 random weights (0.001% of the model) enables hidden skills like sorting and JSON output that prompts alone could not unlock.
A new method specializes Small Language Model vocabularies to reduce measured inference costs by 77% while improving quality scores on domain-specific tasks.
ToolGrad achieves a higher pass rate using an answer-first approach, while small Gemma-3 models demonstrate competitive performance against SoTA models.
Analysis of Gemma-3-12B reveals that neutral context entered before a question changes the model's internal state and induces answer refusal.
NVIDIA has released the Nemotron 3.5 Content Safety model, which supports multimodal input, multiple languages, and custom policy enforcement.
Google has unveiled the Coral Board, on-device AI hardware capable of running Gemma 3 locally.
Anthropic released NLA interpretability research for Gemma 3.
This provides a comparison of open source translation tools that utilize the Gemma 3-based TranslateGemma model.
Olive Young optimized its review theme recommendation system by building a Gemma 3-based sLLM in a constrained environment with just one T4 GPU.
Google has released Gemma 3, a new open model supporting multimodality and a 128k context.