ChatGPT: The potential (and pitfalls) of large language models

What are large language models?

AI applications summarize articles, write stories and engage in long conversations — and large language models (LLMs) do the heavy lifting. An LLM is a deep-learning algorithm that can recognize, summarize, translate, predict and generate text and other content based on knowledge gained from massive data sets. LLMs can teach AIs human languages, understand a wide variety of subject disciplines, write software code and create graphics based on qualitative text descriptions.

What is ChatGPT?

OpenAI’s ChatGPT was released on November 30, 2022. Built on a large language model – and a product of so-called generative/conversational AI – this chatbot allows users to have human-like conversations as it can recognize patterns and create new outputs based on its understanding. The app’s growth has outpaced the likes of TikTok, Instagram and others.

ChatGPT comparison: Time to reach 100 million users

^{Source: Company data, Credit Suisse, CNBC}

Industries set to benefit from ChatGPT

OpenAI’s technology innovations and the progress of generative/conversational AI are transformative and could benefit productivity, cost-cutting and efficiency across sectors. Opportunities to improve productivity seem most apparent in the Information Technology, education, government, and business services industries.

In other industries, ChatGPT may not be a needle-mover today, or even in the next 12 months, but AI and LLMs eventually will have an impact. As an example, healthcare has lagged other industries in the adoption of AI-based tools despite the potential benefits in cost saving and efficiencies. Estimates suggest that 5%-10% of US healthcare spending could be saved with a wider use of AIⁱ.

Challenges of large language models

LLMs have been unleashing significant potential and creating synergies in areas such as search engines, robotics and code generation for some time, and ChatGPT has become the most popular application of LLM. However, LLMs do face some challenges: high costs to maintain and scale; a long time to deploy; low data accessibility; easily over-trained; and lacking experts.

Risks and regulatory concerns related to ChatGPT and AI technologies

There are growing concerns about state-of-the-art AI capabilities and the potential impact on society as a whole.

• Threats from ChatGPT if being used by bad actors. According to researchⁱⁱ, underground hacking communities (e.g., dark web) have been using ChatGPT to develop malicious tools, simply by leveraging ChatGPT’s AI capabilities.
• ChatGPT may be too helpful. New York City’s Department of Education announced a ban on ChatGPT on public school devices and networks after concern about the negative impacts on students’ critical-thinking and problem-solving skills.
• AI chatbots may not understand the question (and may not care either). Given that the latest training data for ChatGPT (GPT-3) is from 2021, it has limited the chatbot’s ability to answer questions related to recent events. Similarly, uncontrolled responses may greatly impact user experiences and usefulness.

Regulators across the world are taking action

The EU passed the Artificial Intelligence Regulation Act in December 2022. This is the first law on AI by a major regulator anywhere, indicating that the EU is leading the way on AI regulations. EU industry chief Thierry Breton has proposed AI rules that aim to deal with the risks and concerns from ChatGPTⁱⁱⁱ. Meanwhile, US regulators are taking action on AI – for example, the AI Bill of Rights^iv – but not on chatbots as yet.

OpenAI’s management has noted on several occasions that governments should be involved sooner rather than later to make sure the societal impact from ChatGPT is controlled and responsible. There is potential to see more proactive regulation given the high levels of media attention surrounding ChatGPT.