网络抓取与生成式人工智能的兴起.pdf

《网络抓取与生成式人工智能的兴起.pdf》由会员分享，可在线阅读，更多相关《网络抓取与生成式人工智能的兴起.pdf（34页珍藏版）》请在三个皮匠报告上搜索。

1、WEB SCRAPING AND THE RISE OF GENERATIVE AIDavid N.Patariu,PLSPartnerThe CISO Law FAyesha BhattiHead of Digital Policy,UK&EU,Center for Data InnovationInformation Technology&Innovation FoundationStephen AlmondExecutive Director,Regulatory RiskUK Information Commissioners OfficeAnnabel Dalby,CIPTSenio

2、r Manager,EMEIA Cyber Security and Data PrivacyEYWELCOME AND INTRODUCTIONSOverview-what is web scraping?Raise your hand if you know what this isAI models require large datasets to learn language,patterns,and decision-making.Web scraping is a primary data source for text,images,and structured informa

3、tion.However,public availability does not mean unrestricted use;legal and ethical considerations must be addressed.The Role of Data in AI Why Scraping Matters?AI models(e.g.,OpenAIs ChatGPT,Googles Gemini)are trained on massive scraped datasets.Examples of scraped sources:Common Crawl,Wikipedia,soci

4、al media platforms,news sites,books,transcripts of video clipsLegal uncertainty and ethical dilemmas emerge as AI development acceleratesThe Scale of Web Scraping for AIWill we run out of data?“If rapid growth in dataset sizes continues,models will utilize the full supply of public human text data a

5、t some point between 2026 and 2032,or one or two years earlier if frontier models are overtrained.At this point,the availability of public human text data may become a limiting factor in further scaling of language models.Performance Plateau in AI Models-AI models are reaching a limit;adding more pu

6、blic data no longer improves performance significantly.Scaling Alone Isnt Enough-Experts,including Ilya Sutskever,suggest that the bigger is better approach is hitting a wall,requiring new strategies beyond just more data and compute.Synthetic Data Has Limitations-AI-generated synthetic data hasnt d