2023/11/24 | Written by: Sungmin Park

If you were to describe 2023 in one word, what would it be? One of the keywords that ignited curiosity throughout the world was Artificial intelligence—so much so that Collins Dictionary, Britain's representative English dictionary, selected 'AI' as the word of the year on the 1st. The frequency of AI terms used has more than quadrupled this year compared to last year. In this era, which feels like the heyday of artificial intelligence, AI technology is developing at an incredible pace, pioneering innovation and change not just in various industries but in our daily lives as well. As the year comes to an end, we take a look back on the major issues in the AI industry that have enriched our society with five key words.

1. AI democratization

AI democratization refers to the phenomenon in which anyone can easily access and utilize artificial intelligence, universally enjoying the benefits of technology. Until Chat GTP familiarized the world with various generative AI services such as AI profile photo creation, artificial intelligence was a field of expertise distant to the public. It was difficult for individuals unexposed to programming languages or for companies lacking in capital, manpower, and infrastructure to develop models to utilize artificial intelligence technology.

However, accessibility has recently improved on all facets, with a focus on the inclusivity of technology that everyone could utilize. From AI development and design to data sets and analysis, alongside related tools and systems, the democratization of AI has begun. Representative examples which have grown in influence across various industries this year include ‘open source’ and ‘no-code/low-code.’

• Open Source:

Software whose source code is open, allowing for anyone to freely review, modify, and distribute it. Each open source’s license determines how and to what extent users can adapt the software.
[ → Learn more about the open source LLM ecosystem and Korean modeling]

• No-Code/Low-Code:

  An approach to make software development easier and faster. ‘No Code’ is the method of developing software without using a programming language, suitable for creating software without any coding expertise. Conversely, 'Low Code' is a method that simplifies software development by requiring programming but minimizing the writing of code.

  
  

The combination of open source and no-code environments creates opportunities for companies and individuals to develop new ideas, innovating through AI, and is expected to positively impact the future by bridging the digital gap.

2. AIaaS

AI as a Service is a cloud-based subscription AI service commonly referred to as AIaaS. Since the COVID-19 pandemic, digital transformation has flourished, with the number of cases of active artificial intelligence technology use rising across industries. AIaaS, which allows easy introduction of the latest AI technology by paying a small usage fee, has garnered attention due to its time reduction and low cost advantages compared to internalizing AI.

In AI development, not just advanced technical expertise but also substantial initial investment is essential. Therefore, for small and medium-sized enterprises (SMEs) and startups with limited resources in terms of manpower and capital, AIaaS has emerged as a groundbreaking solution. By leveraging AIaaS, companies can receive AI technologies settled throughout the cloud in the form of APIs without needing specialized knowledge or workforce. Moreover, if an AI development environment is needed, it is also possible to receive and utilize AI development tools and environments as a service.

Representative examples of AIaaS solutions include the cloud computing service 'MicrosoftS Azure AI,' alongside the 'AWS (Amazon Web Service)' created by Amazon. Furthermore, in Korea, AI OCR is being utilized along with Naver's 'HyperclovaX.' Upstage stands out, being an AI startup that allows users to engage it in an easy and convenient API form. Upstage recently opened the [→ 'Upstage Console' ], allowing users to freely introduce and experience OCR technology, which can extract information from various types of documents such as handwritten, receipts, and even crumpled documents on the console. AIaaS, which allows anyone to smoothly access the latest AI technology, has played a significant role as a catalyst for the digital market this year.

3. Generative AI

Additionally, 2023 marks a significant year in the history of artificial intelligence due to the generative AI boom. Generative AI systems utilize machine learning and natural language processing technologies to create various forms of content such as text, images, and audio based on specific user requirements. The generative AI craze, sparked by ChatGPT, is regarded as a game-changer that introduces new paradigms in human history, akin to the transformative Iron Age. ChatGPT reached 1 million users within 5 days of its launch last November, becoming a phenomenon and achieving a revolutionary milestone even compared to global social media services such as Instagram and Facebook. Besides ChatGPT, which has popularized large language models (LLMs), other models like Meta's 'LLaMA' series, Google's 'LaMDA' and 'PaLM2', and Stanford University's 'Alpaca' are continually improving their performance and addressing weaknesses, thereby expanding their influence as they progress.

[ → A look at the GPT series and its development process ]

Furthermore, services such as 'Midjourney', 'DALL·E', and 'Stable Diffusion' have emerged in the generative AI industry, allowing consumers to create desired images by simply entering a text-based prompt. In particular, AI profile picture creation services have gained significant popularity in South Korea, leading to a surge in inquiries about whether AI-generated profile pictures could replace ID photos and prompting the Ministry of the Interior and Safety to officially announce that such images cannot be used for resident registration cards. This example illustrates just how widespread generative AI has become.

Recently, big tech companies such as Google and OpenAI have announced their focus on developing multimodal AIs, heralding a new sensation in the generative AI market. Multimodal AI systems are evolving to process and integrate various types of information such as text, image, voice, and video simultaneously, providing comprehensive understanding and responses. There is growing attention on what the future holds for generative AI as it advances into multi-modal AI.

4. sLLM (Small Language Model)

In the competition among big tech companies releasing LLMs (Large Language Models), sLLMs (Small Language Models) have shown their small but strong presence. sLLM stands for Small Large Language Model, and ever since Meta allowed academic access to LLaMa, numerous sLLMs utilizing open-source frameworks have emerged and started gaining attention.

Compared to traditional LLMs, sLLMs have fewer parameters but maintain high performance through training on high-quality data and fine-tuning. This allows for a reduction in the required computing resources, leading to significant cost savings. While typical LLMs have over 100 billion parameters, sLLMs highlight the difference, usually consisting of 6 billion (6B) to 10 billion (10B) parameters. Moreover, sLLMs can be designed to prevent issues and concerns for many companies, such as the leaking of confidential information or hallucination phenomena. Due to their low-cost, high-efficiency advantages, the demand for sLLMs is rapidly increasing among businesses looking to effectively apply generative AI to specific industries, products, or services.

[ → View Upstage StarView, which achieved world-class performance with sLLM ]

5. AI TRiSM (AI Trust, Risk, and Security Management)

As artificial intelligence circulates through our daily lives, there has been a growing emphasis from voices concerned about the importance of trust, risk, and security management in AI. AI TRiSM, Artificial Intelligence (AI) Trust, Risk, and Security Management, reflects these issues, emphasizing the need for essential governance to ensure reliability, fairness, and privacy. Gartner, an IT research firm, has also highlighted AI TRiSM as one of this year's technology trends.

For AI TRiSM, the following four elements must be considered:

• Explainability (Explainable AI) : Ensuring that users clearly understand and trust the results and predictions output by the AI model. To achieve this, it is necessary to build trust between service providers and users by establishing a process that can efficiently verify the data sets used to train AI models and the model’s produced results.

• ModelOps (AI Model Operation) : Developing processes and systems for overall management of artificial intelligence models, from development and distribution to maintenance and management. This also includes overseeing the underlying infrastructure environment, such as cloud resources, to ensure that the model runs smoothly.

• AI Application Security : AI models often handle vast amounts of sensitive data, making thorough application security a priority. Security keeps models private and allows organizations to configure security protocols to prevent unauthorized access or the tampering of data.

• Privacy : Protecting the data that serves as the backbone of the AI system is directly related to the safety and reliability of artificial intelligence. It is important for companies to comply with privacy regulations while appropriately collecting and utilizing data needed to train models.

Wave of Change: The Era of AI is Here

Starting with Artificial Intelligence, innovative change is just beginning. As technology develops at a rapid rate, solving AI ethics issues will become the key to opening a safer future. The age of AI is upon us; what will stir the world next year?