2023 was a year when AI technology attracted more attention than ever due to the emergence and spread of generative AI products, including AI chatbot "ChatGPT." ChatGPT, developed by OpenAI, surpassed 100 million monthly active users (MAU) in January 2023, two months after its launch on November 30, 2022. The OpenAI CEO, Sam Altman announced that 92% of the top 500 companies use OpenAI's generative AI products. Since then, competition between AI technology and products has accelerated with the release of GPT-4, a high-performance large-capacity language model (LLM), on March 14, 2023. According to a recent Reuters report, OpenAI's annual revenue has surpassed $1.6 billion, with the increase attributed to robust growth in ChatGPT products. This is a significant increase from $1.3 billion in mid-October.
AI powerhouse Google announced the benchmark score of its next-generation AI model, Gemini Ultra, which surpassed GPT-4 on December 6, making it fierce once again. The tech giant is ready to stage a competition. Gemini Ultra has become the first AI model to outperform human experts in Multi-Task Language Understanding (MMLU), which tests knowledge and problem-solving skills by combining 57 subjects, including mathematics, physics, history, law, medicine and ethics, with a score of 90.0 percent.
Market experts say major technology companies have invested heavily in generative AI research and development to gain a competitive edge and shape the future of the industry. The race for AI supremacy will continue in 2024, making it a pivotal year for AI technology.
What will the major players in the field be pursuing in 2024?
The first is OpenAI, a research organization behind some of the most influential and groundbreaking generative AI models, such as GPT-3, DALL-E, and Codex, which can generate natural language, images, and code, respectively, from user input. OpenAI also provides a commercial API that enables developers and enterprises to access and integrate these models into their applications.
OpenAI GPT-4 is the latest version of the popular natural language generation model, which can generate coherent and varied text from user input. GPT-4 is estimated to have over 500 billion parameters, making it the largest AI model ever trained. GPT-4 can be used for a variety of applications, including content creation, summarization, translation, dialog, and more. However, it also poses ethical and social risks, such as generating misinformation, plagiarism, bias, and addiction. To address concerns about the potential risks associated with generative AI, OpenAI has proactively implemented safeguards, such as establishing a Safety Advisory Group and giving the Board of Directors veto power over AI safety decisions.
The Gemini Ultra Challenge
However, the dominance of ChatGPT and GPT-4 was soon challenged by a new competitor: Gemini Ultra, a next-generation AI model developed by Google, a subsidiary of Alphabet and a leader in AI research and development. Gemini Ultra claimed to be not only a natural language generation model, but also a multimodal model, meaning it could generate content in multiple forms simultaneously, such as images, audio, video, and 3D. Gemini Ultra also claims to be able to learn from any type of data and environment, making it more general and adaptable than ChatGPT and GPT-4.
Second, DeepMind, a subsidiary of Alphabet, focuses on creating general-purpose, adaptive AI systems that can learn from all types of data and environments. DeepMind is known for its achievements in reinforcement learning, such as AlphaGo, AlphaZero, and MuZero, which can master complex games without human guidance. The company is also developing generative AI models such as BigGAN, WaveNet, and GANPaint, which can synthesize realistic images, audio, and art from latent vectors or user input.
Third, NVIDIA has moved quickly with the rise of AI in the areas of graphics processing units (GPUs) and accelerated computing, which are essential to powering AI applications. Nvidia also develops its own generative AI models and frameworks, such as StyleGAN, Riva, and Jarvis, which can create high-quality faces, speech, and conversational agents from random noise or user queries. Nvidia also offers a cloud platform, Nvidia AI Enterprise, that enables enterprises to deploy and manage AI workloads on their infrastructure.
In November of this year, NVIDIA announced the launch of its NVIDIA DGX Cloud First with the NVIDIA GH200 NVL32, which will soon be available on AWS. They also introduced Project Ceiba, a high-speed, GPU-powered AI supercomputer designed for NVIDIA AI research and custom model building.
Fourth, another fast-rising company, Builder.ai, a startup that aims to democratize software development and enable anyone to build custom apps without coding. Builder.ai uses generative AI models, such as Copilot and Copilot in Bing, to help users write, rewrite, improve, or optimize their code based on their specifications and preferences. Builder.ai also offers a platform, Builder Studio, that allows users to design, build, and launch their apps using pre-built components and templates.
Fifth, Amazon Bedrock is a new service that allows users to create and manage generative AI models in the cloud using Amazon's SageMaker platform. Bedrock provides an easy-to-use interface and a library of pre-trained models such as StyleGAN, WaveNet, and BERT, which can generate images, audio, and text, respectively. Bedrock also provides tools for data preparation, model training, deployment, and monitoring. Bedrock can help users accelerate their generative AI projects, but it also requires careful data governance and security practices.
Sixth, Google Vertex AI is a unified platform that integrates Google's various AI products and services, such as AutoML, TensorFlow, BigQuery, and Cloud AI. Vertex AI enables users to build, deploy, and manage end-to-end machine learning pipelines, including generative AI models such as BigGAN, WaveNet, and T5, which can synthesize images, audio, and text, respectively. Vertex AI also provides data labeling, model explainability, and fairness capabilities1. While Vertex AI can simplify and streamline machine learning workflows, it also raises concerns about data privacy and ownership.
Seventh, Salesforce Einstein GPT, a natural language generation model integrated with the Salesforce CRM platform, can help users generate personalized and relevant content for their customers and prospects. Einstein GPT can generate text such as emails, headlines, product descriptions, and reviews based on user input and preferences. Einstein GPT can also learn from user feedback and improve over time. Einstein GPT can improve customer experience and engagement, but it also requires human oversight and quality control.
Eighth, Microsoft Copilot is a code generation model integrated with Microsoft's Visual Studio Code editor that can help users write, rewrite, improve, or optimize their code based on their specifications and preferences. Copilot can generate code snippets, functions, classes, and even entire programs in a variety of programming languages, including Python, JavaScript, and C#. Copilot can also learn from user code and suggest improvements. Copilot can increase developer productivity and creativity, but it also brings challenges such as code quality, security, and licensing.
Gartner recommends that customers conduct their evaluations of the productivity benefits of coding assistants, rather than relying solely on the claims of software vendors. Gartner notes that these wizards have occasionally made mistakes, causing concern among security managers. On the other hand, it advises developers to thoroughly test, review, and verify the code suggested by Copilot.
The future of generative AI in 2024
As generative AI continues to evolve, the capabilities and applications of generative AI products are expected to expand, leading to new and innovative applications across industries. According to analysts at Open AI, here are some possible trends and developments for generative AI in 2024 and beyond.
More than just text: Voice interactions will expand. Generative AI products will generate voice in addition to text, enabling more natural and engaging interactions with users. Voice generation will also improve the accessibility and usability of generative AI products for people with disabilities or language barriers.
Modality will expand: Generative AI products will not only generate content in multiple forms, but also across multiple domains, such as art, music, science, and more. Generative AI products will also be able to combine and transform different modalities, such as converting text to image, image to video, video to music, and more.
AI planners: Generative AI products will generate not only content, but also plans, strategies, and solutions based on user goals and preferences. Generative AI products will also be able to optimize and improve their plans based on feedback and results.
Rise of autonomous agents: Generative AI products will not only generate content and plans but also act on them by interacting with the environment and other agents. Generative AI products will also be able to learn from their actions and experiences and adapt to changing situations and contexts.
Vector Databases and Embeddings: Generative AI products will not only generate content and data but also store and retrieve it using vector databases and embeddings. Vector databases and embeddings are methods of representing and organizing content and data in a high-dimensional space where similar or related items are closer together. This enables faster and more efficient search and retrieval, analysis, and comparison of content and data.
Generative AI services, which are revolutionizing industry and society, have the potential to unleash human creativity and innovation in unprecedented ways, but they also have the potential to have profound social and human impacts that require careful and responsible management. The harms of Generative AI can create harmful or malicious content and data, such as misinformation, plagiarism, bias, and addiction, which can undermine trust, credibility, and quality. It can also raise ethical and social dilemmas, such as who owns, controls, and benefits from the generated content or data, and who is responsible and liable for the consequences. Generative AI can also challenge human identity and agency by blurring the boundaries between human and machine, and by influencing or manipulating human behavior and decision-making.
