In 2023, Generative Artificial Intelligence (AI) came out of the lab and brought us unexpected innovations that piqued the world's curiosity and challenged what we knew about AI until then. ChatGPT, launched at the end of 2022, was the first example, reaching 100 million people in just two months, but there were many other developments during 2023. In addition to Open AI, several startups such as Stability AI, Midjourney, Anthropic, among others, have viewn their valuation multiply over the past year.
On a business level, 2023 was a year with an astonishing pace of innovation. In Europe, more than 2/3 of companies already use AI solutions in their core business, and by 2024 a further 21% will do so too. But the potential of Generative AI is only beginning. The first half of 2023 was fundamentally focused on exploring business use cases, across various industries. In Portugal we had several examples: the Ministry of Justice created a solution based on chatGPT for citizen service, the Agency for Administrative Modernization created an avatar with Generative AI to support citizens in using the digital mobile key, in the private sector, Helena, the chatbot with Generative AI made available by CTT for customer service customer, and Talkdesk which started integrating Azure Open AI services into its CRM.
In 2024 we expect a democratization in the use of Generative AI, both in individual terms with technologies such as Microsoft Copilot, in which these models are fully integrated into the tools we use every day, such as Excel or PowerPoint, and in business terms, with an expansion of use cases to more sophisticated end-to-end solutions, which will allow us to move on from the chatbots we saw emerge in 2023, to more intelligent, hyper-personalised applications with a much more intuitive and contextualized interaction capacity, which will move from text only, to image, voice and video.
This article looks at some of the main trends for 2024 in the area of Generative AI, with a practical perspective on the use of the technology and the impact it will have on companies and all of us as users. Today more than ever, organisations need specialised resources, more comprehensive tools and new skills to create the next wave of intelligent applications based on Generative AI:
1. The democratization and sophistication of Generative AI
After acceleration, democratization. According to Gartner's estimates, more than 80% of companies will have Generative AI models in production by 2026, a quantum leap, since at the start of 2023 the same rate was below 5%.
Democratization will largely be due to the accessibility of the various LLMs, through Cloud services, which provide scalable GPU infrastructures, a critical factor for the deployment of these intensive computing solutions, and through the availability of the LLMs themselves, already pre-trained, as a service (Model-as-a-Service), ready to be integrated into more complex applications and business processes.
Finally, the availability of LLMs in open-source models will also be a democratizing factor, especially in specialised or less demanding scenarios in terms of model accuracy. Examples such as Meta's Llama 2, Microsoft's Phi2 and the hundreds of other open-source models available on Hugging Face allow us to foreview this trend today.
2. Multimodal Generative AI, where voice, image and video will be "kings"
Despite the innovation of foundational models during 2023, the new wave of Generative AI will make it possible to carry out much more complex activities that will be characterized by multimodality, i.e. the ability to understand, process and generate various types of information, including text, voice, image, video, among others. These advanced capabilities open up more creative opportunities and more comprehensive experiences, in which AI will gain a better understanding of the world as we know it.
This trend already took off in 2023, with the inclusion of Open AI's DALLE-3 model in Bing Chat, enabling image analysis and generation, followed by the voice component, and more recently the inclusion of vision in Open AI's GPT-4 models, namely GPT-4 Turbo with Vision. The same trend was followed with Google's recent launch of its Gemini multimodal model, capable of processing images and video.
Multimodal interaction will revolutionise the way we interact with computers, and will be a fundamental pillar for the development of new intelligent applications, which will not only understand us better and more naturally, but above all will extend our capabilities.
3. The age of Copilots and the impact on the new world of work
Another dominant trend for 2024 involves the use of Generative AI in virtual assistants that help us take part in day-to-day tasks, streamline processes, increase efficiency and productivity. A recent Microsoft study on the future of work concludes that we spend 57% of our time on administrative tasks, such as managing emails, videoconferences and chats, and only 43% on creative tasks.
The Copilot concept, which Microsoft announced for 2023 and which, nine months later, has come to fruition with the inclusion of a personal assistant - Microsoft Copilot - in the tools we all use, such as PowerPoint, Excel or Teams, uses Generative AI (in particular the GPT-4 and DALLE-3 models) and applies it to our specific domain of business information, such as documents, emails, meeting recordings, among many others. Copilot represents the new human-machine interface for the knowledge that resides in organisations and will help us act on that knowledge.
The year 2024 will undoubtedly be the year when the flourishing of specialised models, whether embedded in productivity tools or business applications, will have a direct impact on the productivity and satisfaction of all of us, but also on the way we do business in the future.
4. The surprising power of Small Language Models (SLMs)
LLMs such as Open AI's GPT-4, Google's PaLm 2, or Anthropic's Claude, have demonstrated a remarkable ability to reason, answering complex questions, and even solving problems that require multi-step reasoning, capabilities that were previously considered beyond the reach of AI. The challenge is whether such emerging abilities can be achieved in smaller-scale models, using different training strategies, ranging from the selection of high-quality data, the use of synthetic data, to the architectures of the models themselves.
Microsoft Research's launch of the Orca language models (13 billion parameters) and, several months later, Orca 2 (7 billion and 13 billion parameters) demonstrated how improved training methods, such as the creation of synthetic data, can raise the reasoning of SLMs to a level equivalent to that of models 10 times larger. Another example from Meta was the launch in June 2023 of the Llama 2 model, a collection of open-source models that range in scale from 7 to 70 billion parameters and are freely available for commercial and research use. Llama 2 outperforms other open-source language models on many external benchmarks, including reasoning, coding, proficiency and knowledge tests.
Microsoft recently expanded its family of SLMs with the 2.7 billion parameter Phi-2, which raises the bar for reasoning and language understanding between base models with up to 13 billion parameters. Phi-2 has also met or exceeded the performance of models 25 times its size, achieving outstanding performance on a variety of benchmarks, thanks to the use of high-quality, carefully selected data that is filtered on the basis of educational value and content quality.
5. Generative Enterprise AI - from chatbots to intelligent applications
Generative AI will revolutionise the way companies operate. From advanced analytics to the automation of labourious tasks, organisations will measure the impact of AI on the productivity and creativity of their employees. One of the trends pointed out by Gartner is intelligent applications, defined as applications capable of responding appropriately, dynamically and autonomously, and that starts with data. Many organisations already have the data they need to generate meaningful insights with AI, but unless they integrate that data into intelligent applications, they won't be able to take full advantage of it to grow, innovate and provide excellent service.
In 2024, Generative AI will redefine the way companies create business applications. While today most applications that support business are static, with predefined interfaces, programmed for known scenarios, and with high switching costs, the new paradigm will lead us to intelligent applications that use natural language as a means of interaction, with personalised experiences that improve over time and with impressive semantic processing capabilities.
Chatbots may have been the gateway for organisations into the world of Generative AI, but they were only the beginning. With the introduction of Generative AI into business solutions, many applications will be completely redefined.
By Manuel Dias
Microsoft Portugal