Generative artificial intelligence already started nine decades ago and is developing at an incredible pace, with new applications seeing the light on a regular basis.
The remarkable velocity and unprecedented speed is evident from the number of technological breakthroughs following the release of OpenAI’s ChatGPT in 2022. Some of the major large language model (LLM) developments are:
– 2022: Stability AI developed Stable Diffusion, a deep learning text-to-image model that generates images based on text descriptions. This led to the rise of other diffusion-based image generators, such as DALL-E 2 by OpenAI and Midjourney.
– 30.06.22: Google released Minerva, which can solve complex mathematical problems at university level.
– 30.11.22: OpenAI released ChatGPT powered by GPT-3.5 and reached 100 million users in two months.
– 12.12.22: Cohere released the first Large Language Model (LLM) that supports more than 100 languages.
– 26.12.22: LLMs like Google’s Med-PaLM are trained for specific use cases, for example, clinical knowledge. It can generate high-quality text, explain jokes, cause and effect, and more.
– 2023:The generative AI race begins in all earnest.
02.02.23: Amazon incorporates “chain-of-thought (CoT) prompting” in their multi-modal CoT model. This model explains its reasoning and outperformed ChatGPT on several benchmarks.
22.02.23: Microsoft released its ChatGPT-powered Bing search engine chat. Updated to GPT-4 on 14 March 2023.
24.02.23: Meta releases its smaller, more efficient Large Language Model Meta AI (LLaMA).
– 27.02.23: Microsoft releases Kosmos-1, a multi-modal LLM that responds to natural language text, image and audio prompts.
– 07.03.23: Salesforce announces Einstein GPT, based on OpenAI’s model. It is the first generative AI technology for customer relationship management (CRM).
– 13.03.23: OpenAI releases GPT-4 multi-modal LLM that can receive both text and image prompts. GPT-4 came with significant improvements in accuracy and the mitigation of hallucinations.
– 14.03.23: Anthropic launched Claude, an AI assistant that was trained using “constitutional AI” to reduce the likelihood of harmful outputs.
– 16.03.23: Microsoft announces the integration of GPT-4 into its Office 365 suite to enhance productivity.
– 21.03.23: Google released its own generative AI chatbot, Bard, based on its Language Model for Dialogue Applications (LaMDA) engine.
– 30.03.23: Bloomberg announced an LLM trained in financial data for use in the financial industry.
– 13.04.23: Amazon introduced Bedrock, the first fully managed service that makes models available via Application Programming Interfaces (APIs) from Amazon’s Titan LLMs, as well as several other providers (e.g. Anthropic).
– 23.04.23: OpenAI released a beta version of its browser extension for ChatGPT with potentially unbounded access to real-time data on the web, as well as third-party plug-ins.
– 09.05.23: OpenAI launched Shap-E, a tool that can generate 3D models from images or text.
– 07.07.23: Huawei unveiled its self-developed Pangu 3.0 AI model that runs on Ascend AI-powered processors.
The combination of larger labelled data sets, more powerful computers and new ways of automatically encoding unlabelled data has hastened generative AI's development over the past five years. The year 2023 alone has seen remarkable chatbots, numerous new services for generating images from text input, and the adaptation of LLMs to virtually every aspect of a business. One of the major challenges of generative AI is the management of a technology moving at a speed not seen in previous technology transitions.
No wonder many CEOs of companies are struggling to understand AI technology’s business value and risks. They need to determine if this is just tech hype or an opportunity to leapfrog the competition.
The CEO plays a vital role in driving a company's focus on generative AI and need to keep certain strategies in mind on their AI journey.
Preparing for generative AI
We have been using AI for decades (Google search, Amazon Alexa, Netflix recommendations), but since the release of ChatGPT, individuals can directly and creatively use generative AI to rapidly deliver value through the writing of content, developing software code or performing a task. Since generative AI is so accessible and dynamic, it requires an agile mind-set. CEOs need to think carefully about their company’s domain capabilities and differentiated data assets and how they can be leveraged to the benefit of the company and customers.
It is important to create a culture where smart AI experimentation is valued and encouraged. Although many organisations in the past started their AI journey through siloed experiments, generative AI’s ability to underpin multiple uses across the organisation demands a more coordinated approach.
A good start is to convene a cross-functional group of the company’s leaders to identify and prioritise the highest-value use cases but also enable coordinated implementation across the organisation.
New skills required
While generative AI offers greater accessibility compared to traditional AI, it still requires the acquisition and adoption of new skills by the existing workforce to effectively harness generative AI technologies and remain competitive in the market.
Companies will, therefore, need to train and educate their existing workforce. Although prompt-based conversational user interfaces make generative AI applications easy to use, users still need to optimise their prompts, understand the technology’s limitations, and know where and when they can integrate the application into their workflows. CEOs could leverage generative AI to provide personalised and adaptive training materials, simulations, and interactive learning experiences.
Hiring the right talent is also important. To develop generative AI tools using existing models and SaaS offerings, a data engineer and a software engineer will be needed.
The technology stack
A modern data and technology stack is key to a successful approach to generative AI. In addition to adequate computing resources, tools and access to models, real-time access to data is crucial to generative AI. Generative AI relies on large volumes of high-quality data to produce accurate and meaningful outputs. AI can easily analyse massive volumes of data and generate insights and predictions that help businesses make informed decisions. This can lead to more accurate forecasting, optimised supply chain management, improved risk assessment, and enhanced strategic planning.
Equally important is investment in a scalable data architecture that includes data governance and security procedures to maximize the potential of generative AI.
Showcasing
Due to the rapid developments in generative AI, CEOs should not get stuck in the planning phase. The fast-paced nature of generative AI technology demands that companies move quickly to take advantage of it. CEOs should showcase within the company how AI can impact the company’s operating model. The lighthouse approach can be followed by building an application that supports staff with queries and knowledge, and increases productivity and creates enthusiasm. This allows the company to test generative AI internally before rolling it out to customers.
By automating repetitive tasks and enabling innovative approaches, companies can unlock new revenue streams and gain a competitive edge. However, in all cases proof-of-concept is necessary to test and refine a valuable business case before scaling to use cases.
Embracing generative AI
Generative AI presents significant opportunities for CEOs to foster innovation, enhance customer experiences, and drive business value. Embracing generative AI will position businesses for success in the era of AI-driven transformation.
Although the current focus lies on improving productivity, efficiency, cost reduction, and addressing technical limitations, an imminent wave of innovation in business models is poised to revolutionise the landscape.
Professor Louis C H Fourie is an Extraordinary Professor in Information Systems at the University of the Western Cape.
BUSINESS REPORT