Q1. 🤔 Gemini models align with which standard framework concerning safety policies?
- Google’s standard framework
-  জাতিসংঘের Framework
- EU AI Act framework
- OECD framework
Answer: Google’s standard framework
Q2. 🛡️ In Gemini 2.5, what specific area has seen increased training compute allocated to it, allowing for deeper exploration and refinement of model behaviors?
- Supervised Fine-Tuning (SFT)
- Automated Red Teaming (ART)
- Reinforcement Learning (RL)
- Data filtering
Answer: Reinforcement Learning (RL)
Q3. 🔬 According to the Frontier Safety Framework, which specific area triggered 'an alert threshold' for Gemini 2.5 Pro, suggesting that models may reach the CCL (Critical Capability Level) in the foreseeable future?
- Cybersecurity Uplift 1 CCL
- CBRN Uplift Level 1 CCL
- Machine Learning R&D Autonomy Level 1 CCL
- Deceptive Alignment Instrumental Reasoning Level 1 CCL
Answer: Cybersecurity Uplift 1 CCL

Safety and Security in Gemini 2.5

Explorer des sujets liés