AI+企业服务的挑战与创新

近年来,AI技术革新极大地拓展了AI的应用领域和影响力。相较之前的大模型,AI的理解和交互能力大幅提升,然而,AI基于固定模式和套路输出的内容往往差强人意、答非所问;面对to C场景尚且如此,那么to B场景也可想而知。

多视角看大模型安全及实践

随着人工智能领域大模型(large model)的广泛应用,大模型,尤其是大语言模型(large language model,LLM)的安全问题受到了广泛关注.大模型作为一种新兴技术,与之相关的安全态势分析以及安全体系建设均亟待挖掘与探索.从社会关系以及技术应用2个视角,分析了大模型安全的整体趋势.同时,基于大模型自身的特点,梳理了大模型安全能力建设的实践思路,为大模型研发、大模型应用构建提供了安全体系构建的参考方案.介绍的大模型安全能力实践方案包括安全评估基准建设、模型价值观对齐方法、模型线上服务安全系统建设3个部分.

企服大模型,加速AI场景落地

2022年年底,OpenAI推出ChatGPT引发全球热潮,加速AIGC技术和应用的发展,标志着人工智能进入普及应用阶段。AIGC作为人工智能技术的重要组成部分,在商业社会中扮演着越来越重要的角色。在诸多AIGC工具中,基于GPT的应用可以让用户通过自然语言快速生成应用和内容。

基于端云协同体系的预训练大模型及其服务化

传统云计算模式将所有数据集中,以中心化原则在云上训练大模型,通过云服务支撑端侧多样服务需求,这一模式存在网络时延大、隐私安全低和算力成本高等不足。在“泛在互联、移动优先、AI赋能”时代,需要机器学习支撑丰富多样端侧应用,因此建立端云协同计算范式,既提供云上服务和端侧推理能力,又推动云上模型和端侧模型的协同进化,从云计算和端智能向端云协同进化计算模式进行跨越,充分发挥云上、端侧和端云链中各类计算资源的最佳效果是当前的重大挑战。本文围绕云上预训练大模型、端云协同体系、基于端云协同体系的预训练大模型服务化与隐私保护,以及未来挑战进行了讨论。

数智化驱动高质量发展

企业实现高质量发展的路径之一,就是要推进数智化转型,以数字化为基础,智能化产生新价值,最终成为数智企业。数智企业应具备六个特征:客户导向、生态共荣、员工能动、实时感知、数据驱动和智能运营。成为数智企业是为了创造更大价值,使企业有更高的经营绩效、更强的竞争优势,实现可持续高质量发展。

Discussion and Reflections on University Library of Subject Knowledge-based Service Model

With the rapid development of computer technology and network technology, the service model of university library gradually goes to the intelligent, personalized directionfrom the traditional single reference, but there are many shortcomings. It cannot fully satisfy the increasing needs for the division Health research and information, it must actively build the University Library subject Knowledge service Model of a science. This paper introduces the subject of service concepts of library knowledge, proposes effective method to build service model of university library knowledge, and strengthens academic knowledge service countermeasures to promote the development of knowledge- disciplinary university library services.

Detector control system for Daya Bay Reactor Neutrino Experiment

The Daya Bay Reactor Neutrino Experiment is to measure the smallest mixing angle θ13.The experiment contains three major experiment halls,Daya Bay near site,Linao near site and far site,and two major kinds of detectors,antineutrino detector which is to detect the antineutrinos by the inverse beta-decay reaction in Gd-LS,and muon detector which is to study and reject cosmogenic backgrounds.The goal of the detector control system(DCS)is to operate and detect the detectors and keep them running in safety.In consideration of the limited fund of this system and manpower of working on this system,the LabVIEW is chosen to develop the detector control system.The architecture of DCS adopts the distributed data management which is based on client-server model.The server part is to detect and operate parameters from hardware,save data to database and release data to clients,the client is to receive data from the server.The detector control system contains three parts:the hardware part,the local control system and the global control part.The local control system includes high voltage supply system,low voltage supply system,VME crate system,temperature and humidity system,gas pressure system,and so on.

NIBU:An Integrated Framework for Representing the Relation Among Building Structure and Interior Utilities in Micro-Scale Environment

This paper describes a framework for modeling interdependencies between different network systems and building structures.It provides an approach for the integrated analysis of interior building utilities by describing a framework to model and simulate infrastructure interdependencies and their complex behaviors.It is a graph-based spatial model that can support use cases such as providing the location and specifications of interior utilities to a technician who wants to perform a maintenance operation.This location could be needed for maintenance or replacement,or to investigate the result of damage to the building structure on another utility network,or to estimate the effect of different maintenance operations in different locations along utilities service systems.The model accounts for two important aspects:first,the relationship between interior utilities and building elements or spaces and second,the building hierarchy structure to which the utilities network is related.A proper hierarchy of the building is developed which supports the generation of human-oriented descriptions of interior utilities,where a method for partitions of large building element and spaces as well as a method to reference a network element to another building are developed.The connection of the different utilities network systems and buildings are generated using joints,which are based on a containment relation.An example is presented which shows the effectiveness of this approach for supporting maintenance operations,as well as the independences between the maintenance operation location and the other network systems.The paper presents the data model and explains the links with current 3D building model standards.