2朱小虎Responsible AI Interpretability Privacy Governance.pdf

1、Responsible AI:Interpretability Privacy GovernanceShanghaiResponsible AIShanghaiResponsible AIShanghaiGeminiGeminis capabilities and potential applications are vast,but addressing ethical considerations and ensuring responsible development will be crucial for its success.Gemini is natively multimoda

2、l,which gives you the potential to transform any type of input into any type of output,reasoning seamlessly across text,images,video,audio,and code.Gemini“Were approaching this work boldly and responsibly.foundation models and infrastructure and bring them to our products and to others,guided by our

3、 AI Principles.”-Sundar Pichai“At Google,were committed to advancing bold and responsible AI in everything we do.Building upon Googles AI Principles and the robust safety policies across our products,were adding new protections to account for Geminis multimodal capabilities.At each stage of developm

4、ent,were considering potential risks and working to test and mitigate them.”-Demis HassabisGemini 1.01.Gemini Ultra our largest and most capable model for highly complex tasks.2.Gemini Pro our best model for scaling across a wide range of tasks.3.Gemini Nano our most efficient model for on-device ta

5、sks.Gemini 1.0“Using a specialized version of Gemini,we created a more advanced code generation system,AlphaCode 2,which excels at solving competitive programming problems that go beyond coding to involve complex math and theoretical computer science.-Demis HassabisAlphaCode 2Figure 1|High-level ove

6、rview of the AlphaCode 2 system.Gemini ProBard use Gemini ProBefore bringing it to the public,we ran Gemini Pro through a number of industry-standard benchmarks.In six out of eight benchmarks,Gemini Pro outperformed GPT-3.5,including in MMLU(Massive Multitask Language Understanding),one of the key l