Artificial intelligence (AI) has been in development for many years, but lately, a generative AI-based chatbot called "ChatGPT" has gained widespread popularity worldwide. Released by OpenAI last November, it acquired billions of subscribers in just two months, making generative AI the new darling of Wall Street. With many companies joining the race to release their own models, the virality around ChatGPT is reminiscent of the excitement around other new technology releases such as autonomous driving, Metaverse, and crypto. But is this just another hype, or is it a critical turning point in the advancement of AI? More importantly, how will the generative AI spark innovations in different industries and benefit our society?
To explore these questions, TSVC, an early-stage venture capital fund specializing in investing in deep tech, organized a forum discussion with AI researchers, industry leaders, and educators. They shared their visions on why ChatGPT has captured the world's attention, what it means for the future of AI, and the emerging innovation opportunities in the ecosystem.
Over the years, many research institutions and companies have invested heavily in developing natural language processing (NLP) models. One of the most prominent models is the Generative Pre-trained Transformer (GPT), first introduced by OpenAI in 2018. Building on the success of its previous versions, GPT-1 and GPT-2, OpenAI released GPT-3 in 2020. With a staggering 175 billion parameters, GPT-3 is one of the largest language models (LLMs) available today.
What sets ChatGPT from its predecessors are its contextual understanding, human-like intelligence, and generative capabilities. ChatGPT can engage in conversations, craft stories, and even write poems, showcasing impressive yet sometimes unsettling capabilities. Unlike previous AI models that were limited to analyzing and classifying existing content, ChatGPT is a generative model that can produce entirely new content, not limited to text, but audio, video, music and arts.
OpenAI is not the only company making heavy investment in GPT development. Other notable players include Google, Microsoft, Facebook, and Nvidia, all of whom are working on releases of larger models to achieve higher performance.
While artificial intelligence (AI) has been in development for many years, the recent success of ChatGPT has accelerated the flywheel of generative AI. Its popularity has demonstrated the power and potential of generative AI, acting as a catalyst for further development in the ecosystem. Experts predict that the generative AI market will grow from $8 billion in 2022 to over $110 billion within a decade, at an annual growth rate of 35%. This flywheel effect will drive the development of higher quality training data, more efficient language models, killer applications, and advancements in hardware and infrastructure resources.
Every innovation comes with risks. The bias, misinformation, deepfake, hallucination, and cyber attacks brought by generative AI have already raised concerns, on the other side, bringing opportunities to startups to provide innovative solutions. We are excited to see ambitious founders adapt to this new trend, innovate and develop technologies in following fields in the coming years.
Sticky applications will drive the flywheel: The success of ChatGPT has inspired companies to explore practical applications that can deliver great user experiences and retain users over time. The accessibility of GPT models through APIs has allowed startups to leverage the model and focus on application development. In coming years, we are likely to see rapid transformations in fields such as chatbot development, content creation, personalized marketing and advertising, and virtual assistants, as generative AI enables more friendly and natural interactions. The ability of AI to generate creative content will also reshape the entertainment and media industries by providing personalized content for each individual.
Meanwhile, the increasing adoption of generative AI will drive innovation in lagging industries with more regulations and proprietary knowledge, such as healthcare and financial institutions, leading to the development of new businesses. In healthcare, for example, AI can help with medical diagnosis and drug discovery, making treatments more effective and personalized.
However, as with any new technology, much remains unknown in the uncharted territories. To unlock the potential of generative AI, entrepreneurs must carefully evaluate customer needs, the limitations of the technology, and how to best capitalize on their investment.
Data is “king”: The saying "garbage in, garbage out" is especially true when it comes to training data quality. Poor quality data or the lack of data can result in biased, plagiarized, or inaccurate models. This is especially concerning in mission-critical areas such as academia, healthcare, and finance, where even the smallest misstep could have catastrophic consequence. For startups, this presents both a challenge and an opportunity. While obtaining high-quality training data can be difficult and costly, startups that invest in collecting and curating such data, have deep understanding of the domain and ecosystem can gain competitive edge.
Moreover, startups that specialize in data collection and curation can also become valuable partners for other companies and organizations looking to develop AI models. This is particular true to the industries with proprietary knowledges, like healthcare, finance. By providing high-quality training data, startups can help to improve LLM performance, driving innovation in various industries.
Cost-down vertical models lower the entry barrier: The development and deployment of LLMs require enormous amounts of training data and computing power, with typical training costs ranging from 2 to 12 million USD per model. The steep cost presents a barrier for many small players in the field. However, startups can still find edges by developing smaller, vertical models, particularly in the industries where required training data are deep rooted from expert knowledge and historical facts.
With properly curated subject-specific data and fine-tuning by subject-matter experts, these cost-down models can achieve comparable performance to larger models for targeted use cases. By specializing in specific verticals and use cases, startups can offer more customized and effective solutions, while also reducing the overall cost and complexity of developing and deploying AI models.
Hardware demand will surge: Massive computing power required in LLM has created opportunities for AI chip companies, which will also drive an upgrade cycle for networking and data storage hardware. While big companies take a prime position to take advantage of this trend, startups can also benefit from the surge in demand, with advantages of being able to adapt to market faster and provide lower-cost, more tailored solutions in specialized areas. We will see a more diverse landscape, rather than a winner-takes-all market.
It is important for companies to stay ahead of the technology curve, adapt to changing market demands quickly, and establish themselves as leaders in this rapidly evolving field.
How to patrol: The growing use of AI tools has sparked concerns about their potential for malicious use, cyber attacks, and the spread of false information through ChatGPT hallucination. The consequences of these risks could be catastrophic, especially in mission-critical applications such as medical solutions or financial questions. As a result, a couple of large banks have announced that they ban their employees from using ChatGPT due to concerns around accuracy, confidentiality, security, and compliance. To address these concerns, we need to see technological innovations emerging beside global regulations. Companies that can provide products or services to solve these challenging problems will be highly sought after.
Despite we are incredibly optimistic about the future of generative AI, we foresee the initial public exhilaration around the technology will settle as practical limitations and capitalization issues arise, just like many other innovations in past. It will take many iterations to reach its full potential. However, ChatGPT has accelerated the flywheel, calling innovations in various industries in the ecosystem. The startups who can innovate, identify niche markets, and focus on ROI will ultimately emerge as winners in this space. TSVC has specialized in investing early-stage companies in deep-tech and already made a number of investments in this field. If you are a founder and would like to meet, please contact us at email@example.com.