学术报告
【学术报告】(线上)Large language models and ChatGPT
编辑:魏佳发布时间:2023年03月31日

报告人:任伟(微软亚洲软件研究院)

时  间:202345日上午10:00-11:30

地  点:腾讯会议ID311-854-347(无密码)

内容摘要:

ChatGPT is a popular chatbot worldwide with over 100 million users, achieving the fastest record of reaching this milestone in history. This achievement is mainly attributed to the fine research and innovative development in the fields of language modeling and natural language processing. This report will delve into the development process of GPT 1/2/3/4 technologies and demonstrate how the potential value of general artificial intelligence was discovered during this process. In order to better serve humans with large-scale language models, OpenAI has adopted a reinforcement learning method based on manual labeling, which significantly improves the product's usability, authenticity, and security. The emergence of GPT-4 has significantly improved the model's capabilities, but it has also raised concerns about the safety of artificial intelligence and its potential impact on future society. This report will explore these issues in depth and provide some thoughts and suggestions.

人简介:

任伟,2009年博士毕业于中科院数学与系统科学研究院,研究图论以及复杂网络。毕业后在微软亚洲软件研究院工作至今,现任首席软件工程师经理(Principle Software Engineer Manager),整体负责微软全球广告的转化率模型预估业务,是大数据处理领域的资深专家。对于计算广告中的点击率预测、转化模型预估应用等有着丰富的行业经验,对于工业大规模的深度学习、特征工程、线上服务底层实现等具有深厚了解。


联系人:宋翀