According to the current development trend of AI technology, AI may disrupt the future of many industries.
Current breakthroughs in AI technology have brought public attention to a specific type of AI — generative AI. Generative AI revolves around analytics, automation, and content generation, both in quality and in quantity.
It is imperative to understand how generative AI fits into practical applications. According to BCG's blog, the field of generative AI is expected to capture 30 percent of the overall AI market by 2025, equivalent to $60 billion.
How does generative AI work?
Generative AI is the part of machine learning that uses neural networks to generate new content. Unlike other AI systems that are programmed to perform specific tasks, generative AI operates on large datasets and generates novel, unique content.
One of the most popular types of generative AI is generative adversarial networks (GANs). GANs consist of two neural networks: a generator and a discriminator. The generator creates new content and the discriminator evaluates the authenticity of the content.
Generative AI has the potential to change the way we apply AI, from generating realistic synthetic data for training AI models, to curating tailored content for customers. The quality of content generated by GANs has improved over time. Today, GAN-generated pictures and videos are virtually indistinguishable from the originals.
Companies like H&M and Nike are already using generative AI to design clothing in an effort to increase productivity and reduce costs. Thanks to AI technology that creates virtual fashion shows, designers can present their creations in a virtual environment. According to a 2022 McKinsey survey, the use of AI has doubled over the past five years, and investment in AI is expanding rapidly.
Generative AI tools like ChatGPT and DALL-E (see below) are enough to challenge existing job roles.
ChatGPT is a big innovation in the AI industry. It is an effective generative AI language model developed by OpenAI, which can quickly respond to user commands and generate content. Based on Reinforcement Learning with Human Feedback (RLHF) technology, ChatGPT can be used in various applications such as customer support, content generation, data analysis, virtual assistants, language translation, etc. At the time of writing ChatGPT runs on the GPT-3.5 language model (a model created using a large amount of data collected from numerous sources).
On the other hand, DALL-E is an AI model developed by OpenAI that combines advanced deep learning techniques such as transformer networks and GANs. This innovative technology can understand and interpret natural language input and even generate images based on text descriptions.
Application Examples of Generative AI
The application of ChatGPT and DALL-E in real-world scenarios greatly improves efficiency and empowers more creativity. Big companies like Microsoft and Google have incorporated ChatGPT into their customer support systems to provide instant assistance to customers.
Furniture retailer IKEA uses AI to create 3D models of its products, allowing customers to preview furniture in their homes. In addition, automaker Lexus used AI to complete a hyper-realistic car design, demonstrating the technology's ability to facilitate innovative design, highlighting the potential of generative AI technology.
Generative AI is touching NFTs (such as brands and media using NFT art), blockchain games (such as creation using asset generation, narrative and story design, and avatar modeling), metaverses (using 3D ecosystems, multi-asset and texture generation) and Web3 development (such as code generation, audit debugging, and workflow automation) areas.
Generative AI tools in Web3 revolutionized the online search sector. ChatGPT's latest integration with Microsoft's Bing provides an enhanced, user-friendly chat interface. Additionally, Generative AI has entered the Web3 realm with its AI Cloud, helping people filter data on the web and streamline SEO content when making web search queries.
Generative AI learns by training on large amounts of data and learning to mimic patterns in that data and has a wide range of applications.
ChatGPT is trained on vast amounts of text from the internet, enabling it to mimic human conversation. It generates images based on textual instructions by learning images and associated descriptions collected from the web.
Stable Diffusion (a generative AI model that focuses on generating images) uses the process of diffusion to create new images by simulating a random walk from a noisy image to a target image. The model is trained to denoise and reconstruct images during training, learning patterns and features that can generate novel images by reversing the diffusion process. There are applications in art, design, advertising and entertainment.
DALL-E，它是 OpenAI 开发的另一种生成式 AI 模型，基于 GPT-3 架构的修改版本，在大型图像数据集及其相应的文本描述上进行训练。DALL-E 可以根据输入文本生成视觉连贯且上下文相关的图像，在艺术、设计、广告和视觉叙事方面具有潜在应用。
另外值得一提的是 Lens Studio，它本身不是生成式 AI 模型，但它使开发人员能够使用计算机视觉、机器学习和其他 AI 技术创建 AR 内容。Lens Studio 是由 Snap Inc. 开发的桌面应用程序，它允许用户在 Snapchat 上创建和发布称为 Lenses 的增强现实 （AR） 体验，它可以帮助为用户创建引人入胜的交互式 AR 体验，并应用于娱乐、营销和教育领域。
而使用生成式 AI 文本工具，可以简化和创新动态游戏元素，比如对话和头像。生成式 AI 还支持 NFT 艺术生成。在 AI 工具上输入一组规则（例如颜色范围和图案），AI 通过随机和数据迭代，在规定的框架内生成艺术品。
这些模型由公司构建和完善，这些公司通过收集更多训练数据并改进模型。像 OpenAI 和 Stability.AI 这样的公司通过收取其技术的使用费用或为个体企业创建专属内容来获利。
生成式 AI 的潜在风险以及应对方法
每个硬币都有正反两面，生成式 AI 也有一些风险，正是因为它拥有接收外界并学习的特性，在不停优化的同时也不可避免地会被一些恶意人士“教坏”。
生成式 AI 的发展使得 AI 造假之风却愈演愈烈。例如，利用 AI 撰写未经证实的文章或完成学术论文，“一键换脸”、“视频合成”等。利用 AI 制造谣言，扰乱网络传播秩序。
4月25日，国内就发生了一起利用 AI 技术炮制虚假不实信息的案件，一男子利用 AI 技术撰写“今晨甘肃一火车撞上修路工人致9人死亡”的不实文章，文章点击量高达1.5万余次。
那么，针对 AI 存在的诸多风险，我们在应用时首先要了解其风险以及应对策略。
风险
知识产权侵权和内容版权问题
AI 生成的内容的质量和真实性
新区块链运行时生成中的架构障碍
基于敏感数据的内容的隐私问题
生成式 AI 的恶意运行
恶意的算法数据输出
应对策略：
使用基于 AI 的内容审核工具，例如 Google 的 Perspective API 或 Two Hat 的 Community Sift
联邦学习、同态加密和匿名化等数据隐私保护技术
使用用于训练 ImageNet、MNIST 等可信度生成 AI 算法的代表性数据集
使用基于 AI 的欺诈检测工具，例如 http://Fraud.Net、Kount、NICE Actimize
AI 内容分析指标，如公平性和问责性指标
制定在 Web3 中使用生成式 AI 的标准
小结
生成式 AI 的自动化为数据计算提供了动力，帮助 Web3 组织将机器学习集成到他们的运营中，这是一个革命性的领域，生成式 AI 正在创新金融，科技、体育、游戏、医疗保健等行业。
ChatGPT，DALL-E 等生成式 AI 正在改变我们创建内容和与内容交互的方式，使用数据驱动的方法生成新的文本、图像和体验。生成式 AI 将融入到各行各业，改变原有的工作角色，颠覆行业的未来。
来源丨CoinTelegraph
声明：NFT中文社区编辑作品，内容仅代表作者立场，且不构成投资建议，请谨慎对待，如文章/素材有侵权，请联系官方客服处理。