【会议演讲PPT】Gartner+开始您的生成式+AI+之旅（APAC）-49页_2mb

报告摘要

Introduction

Gartner provides actionable insights, tools, and guidance to address organizations' critical priorities through webinars and other resources.
The webinar "Get Started on Your Generative AI Journey (APAC)" presented by Albert Gauthier, Sr. Director Analyst, focuses on Generative AI trends and best practices.

What is Artificial Intelligence

AI in 2023 refers to a set of tools based on decades-old statistical analysis augmented by improving computing capabilities.
Classification of AI use cases includes NLP, regression, probability, neural networks, machine learning, and stochastic models.

Distinctions and Utopian/Dystopian Views

Differing views of AI's potential: Optimistic uses include smart city improvements, while darker views highlight societal risks.

AI and Generative AI Overview

Generative AI revolutionizes sectors like smart cities by offering new insights into behavior and patterns.
Gartner defines Generative AI as creating new content, strategies, and methods using large data sets. It is distinguished from traditional AI, which interprets and automates but does not create.

ChatGPT and LLMs

ChatGPT is a large language model (LLM) built on stochastic models, which predict outcomes based on statistical probabilities.
Key capabilities include content creation, question answering, language translation, and code generation.
Limitations: Not sentient, potentially hallucinatory, requiring expert review for reliability, and not universally applicable.

Enterprise Use Cases

Suitable enterprise use cases include Customer Service, Sales & Marketing, HR, Legal, and Software Programming.
Gartner advises practical application aligned with specific business problems.

Foundation Models and Risks

Foundation models represent a significant scaling in AI, enabling few-shot and zero-shot learning.
Organizations must avoid jumping into GenAI without clear use cases, as this may result in pilot failures.

Getting Started in Gen AI Pilots

Successful pilots focus on business potential, not technical feasibility.
Key considerations are identifying impactful use cases, selecting the right technology options, and ensuring alignment with defined goals and risk mitigations.

Deployment Approaches

Out-of-the-box Models: Quick to deploy but limited by lack of customization and integration leverage.
Prompt Engineering: Customizing inputs without altering the model, increasing versatility but requiring new skills.
Custom Model Deployment: Involves extensive customization, training, and monitoring, suited for sophisticated enterprises.

Embedding LLMs in Applications

Different design patterns exist: As-Is, LLM in Application Workflow, or LLM as Secondary Agent.
Recommendations include embedding LLMs within application architectures, ensuring proper security and user interface integration.

Upskilling and Governance

Organizations should develop cross-functional teams with prompt engineering, knowledge graph, and LLMOps skills.
Responsible AI practices include mitigating hallucinations, ensuring transparency, and establishing feedback loops.

Future of Generative AI

The future points toward more embedded, versatile AI with emerging multimodal models and evolving business models.
Current benefits include enhanced efficiency, but concerns about safety, veracity, and ethical deployment remain.

Gartner Positions

Gartner emphasizes moving beyond hype, following the Gartner Hype Cycle, and focusing on the real "Art of the Possible" in AI implementation.

Recommendations and Next Steps

Prioritize business outcomes and implement at scale only after validating through MVPs.
Leaders should encourage a culture open to advancements in AI, supported by clear governance and training frameworks.

展开完整摘要