{"resolved":true,"match_type":"uuid","source_type":"original_book","book_id":"07d591c2-0f4c-45ac-a1d7-103c4807db8f","title":"数字经济：内涵与路径","author":null,"edition":null,"publisher":null,"year":null,"isbn":"9787521745023","requested_by":"printed_page","book":{"source_type":"original_book","book_id":"07d591c2-0f4c-45ac-a1d7-103c4807db8f","id":"07d591c2-0f4c-45ac-a1d7-103c4807db8f","uuid":"07d591c2-0f4c-45ac-a1d7-103c4807db8f","title":"数字经济：内涵与路径","normalized_title":"数字经济内涵与路径","author":null,"edition":null,"publisher":null,"year":null,"isbn":"9787521745023","isbns":["9787521745023"],"parsed":true,"ai_readable":true,"document_type":"book","page_count":361,"navigation_url":"https://workshop.closeclaw.top/ai/book/07d591c2-0f4c-45ac-a1d7-103c4807db8f","title_url":"https://workshop.closeclaw.top/ai/title/%E6%95%B0%E5%AD%97%E7%BB%8F%E6%B5%8E%EF%BC%9A%E5%86%85%E6%B6%B5%E4%B8%8E%E8%B7%AF%E5%BE%84","fixed_page_url_template":"https://workshop.closeclaw.top/ai/book/07d591c2-0f4c-45ac-a1d7-103c4807db8f/page/{printed_page}","path_api_url_template":"https://workshop.closeclaw.top/api/ai/book/07d591c2-0f4c-45ac-a1d7-103c4807db8f/page/{printed_page}","read_url":"https://workshop.closeclaw.top/api/ai/read/07d591c2-0f4c-45ac-a1d7-103c4807db8f","page_url":"https://workshop.closeclaw.top/api/ai/books/07d591c2-0f4c-45ac-a1d7-103c4807db8f/page","search_url":"https://workshop.closeclaw.top/api/ai/books/07d591c2-0f4c-45ac-a1d7-103c4807db8f/search"},"page":{"printed_page":304,"pdf_page":303,"physical_index":303,"physical_index_base":0,"printed_page_status":"recognized","page_sha256":"fb4a46b679cd663267554f4f8f465c723e8feb88336312466128c1411d15c01c"},"chapter":"第八章","content":"预训练大模型（简称“大模型”）。作为当前人工智能发展的重要方向，预训练大模型具有效果好、泛化能力强的特点，进一步增强了人工智能的通用性，让广大开发者可以更低成本、更低门槛地开发AI模型，促进人工智能的广泛应用，成为人工智能技术和应用的新基座。\n\n百度自2019年开始深耕预训练模型研发，先后发布知识增强文心（ERNIE）系列模型。文心系列模型基于持续学习的语义理解框架，从大规模知识和海量数据中融合学习，效率更高，效果更好。2019年12月，ERNIE 2.0以9个任务平均得分首次突破90大关的成绩登顶国际权威排行榜GLUE榜首。2021年7月，ERNIE 3.0英文模型在国际权威的复杂语言理解任务评测SuperGLUE上超越谷歌的T5、OpenAI（人工智能非营利组织）的GPT-3等大模型，以超越人类水平0.8个百分点的成绩登顶榜首。2021年12月，百度与鹏城实验室联合发布全球首个知识增强千亿大模型鹏城—百度·文心，参数规模达到2 600亿，在60多项NLP（自然语言处理）任务中取得最好效果。2022年5月的WaveSummit深度学习开发者峰会上，百度发布10个大模型，包括融合学习任务知识的知识增强千亿大模型、多任务统一学习的视觉大模型、跨模态大模型、生物计算大模型、行业大模型等，并提出支撑大模型产业落地的3个关键路径：\n\n304","previous_page":303,"next_page":305}