2023-05-29-Gartner-超越炒作_ChatGPT+和生成式+AI+的企业影响_25页_926kb

报告摘要

Gartner Webinars Summary: Enterprise Impact of ChatGPT and Generative AI

Core Content

Gartner's webinar "Beyond the Hype: Enterprise Impact of ChatGPT and Generative AI" explores the transformative potential and challenges of generative AI in enterprise settings. It outlines key definitions, use cases, vendor landscapes, and future trajectories for AI adoption.

Key Definitions

Generative AI: AI techniques that learn from data and generate new, original artifacts that resemble the original data.
Foundation Models: Large machine learning models trained on unlabeled data using transformer algorithms, adaptable to various applications.
Large Language Models (LLMs): A subset of foundation models focused on natural language processing.
ChatGPT: A conversational application built on top of an LLM, specifically OpenAI's GPT model.

What Can Generative AI Generate?

Generative AI can be used in several areas, including:

Content Creation & Augmentation: Generating drafts, modifying tone, and creating titles or outlines.
Q&A and Discovery: Finding answers based on data and prompts.
Summarization: Shortening text and converting it to bullet points.
Simplification: Extracting key content from complex material.
Classification: Categorizing content based on sentiment, topic, or intent.

Benefits & Risks of Foundation Models

Benefits:

Versatility: Can be applied across a wide range of tasks.
Lower Cost of Entry: Accessible to organizations with varying budgets.
Accessibility: Easy to integrate and use.
Ecosystem: Supports a variety of applications and tools.
Domain Adaptation: Models can be fine-tuned for specific industries or tasks.

Risks:

Hallucination: Generating false or misleading information.
Copyright Issues: Potential legal challenges with generated content.
Potential for Misuse: Risk of abuse in sensitive or high-stakes environments.
Concentration of Power: Dependence on a few dominant AI providers.
Opacity (Black Box): Lack of transparency in model decision-making.

Deployment Approaches for GPT & ChatGPT

Out of the Box Model Usage: Use ChatGPT as-is with minimal investment.
- Pros: Fast to market, low investment.
- Cons: Limited customization and control.
Prompt Engineering: Use tools to create, tune, and evaluate prompts.
- Pros: Better targeted results, low startup costs.
- Cons: Requires integration with business systems.
Custom Models: Build or license GPT or other LLMs directly.
- Pros: Customization and optimization.
- Cons: High investment and requires specialized skills.

Vendor Landscape

Proprietary Foundation Models:

OpenAI
Google AI
Microsoft
Cohere
Anthropic
AI21 Labs
Alibaba Group
Baidu
Tencent

Open Source Foundation Models:

Stability AI
Eleuther AI
Meta
Hugging Face
Databricks
Zhipu AI
DeepMind
DistilBERT
XLNet

Applications by Vendor:

Content Creation: Jasper AI, Writesonic, Rytr
Knowledge Management: Sana, Algolia, Glean
Workforce Productivity: Supernormal, Cogram
Metaverse: Replikr, Tavus
Software Engineering: GitHub, Tabnine, Replit
Biotech: Insilico Medicine, Exscientia

Enterprise Trajectories

Bern Elliot's Predictions:

By 2025: 30% of enterprises will implement AI-augmented development and testing strategies.
By 2026: Generative design AI will automate 60% of the design effort for websites and mobile apps.
By 2026: The role of design strategist (blending designer and developer roles) will lead 50% of digital product creation teams.

Erick Brethenoux's Predictions:

Revenge of the Software Grease Monkeys: The use of foundation models will shift focus from AI experts to traditional software engineers.
Adapter Models Explosion: Composite AI models will increase by an order of magnitude in the next 3 years.
By Mid-2024: Decision Intelligence will surpass the hype of generative AI.

Frances Karamouzis's Predictions:

1% of Code Delivers 80% of Net New Value: Highlighting the efficiency of AI in software development.
By 2026: Over 100 million humans will work with robotic colleagues (synthetic virtual colleagues) in enterprise settings.

Prompt Engineering & Fusion Teams

Prompt Engineering: Involves creating and refining prompts to get better results from LLMs.
Fusion Teams: Blending citizens (non-technical users) and professionals (technical experts) to leverage both prompt engineering and model fine-tuning skills.