生成式AI是指能够创建新内容的AI系统——文本、图像、音频、代码、视频——而不仅仅是分析或分类现有数据。它是最近AI热潮(ChatGPT、图像生成器)背后的技术,主要由LLM和扩散模型等大型模型驱动。
什么是生成式AI
GENERATIVE AI → AI that GENERATES new content (vs analyzing/classifying existing data):
→ TEXT → LLMs (ChatGPT, Claude) → write, converse, summarize, code
→ IMAGES → diffusion models (DALL-E, Midjourney, Stable Diffusion) → create images from text
→ also: AUDIO/music, VIDEO, CODE, 3D, etc.
→ produces novel, original-seeming outputs (learned from training data patterns)
生成式AI与判别式/传统AI的对比
TRADITIONAL/discriminative AI → ANALYZE, classify, predict on existing data:
→ e.g. is this spam? what's in this image? predict the price
GENERATIVE AI → CREATE new content:
→ e.g. write an essay, generate an image, compose music, produce code
→ generative AI's ability to CREATE is what drove the recent AI excitement/boom
