Making Sense of Generative AI in 2023
What is Generative AI?
Since its launch in November 2022, Chat-GPT, a Generative AI model, has garnered significant attention worldwide. Generative AI, a subset of Artificial Intelligence, utilizes foundational models to generate novel content such as images, text, and audio. This approach is increasingly replacing task-specific models that are commonly used in applications such as code completion and smart assistants. This trend is not a recent development, as the Center for Research on Foundation Models (CRFM) was established at Stanford in 2021 to further advance this area. However, it was not until the launch of Chat-GPT that the general public became aware of the potential for consumer-facing products utilizing these models. This article aims to provide personal insights into the current market trends of Generative AI.
Market Size and Growth
During my undergraduate studies at Villanova University, while working on a senior project involving an image-to-text model application, I was first introduced to the field of AI. At that time, in 2015, the global AI market size was relatively small at $126 billion, as compared to a projected size of approximately $3 trillion in 2024, with a Compound Annual Growth Rate (CAGR) of 37.3% from 2023 to 2030.
As per NFX’s estimation, the number of VC-backed startups specializing in Generative technology has now exceeded 500. In the years 2021 and 2022, the trend of increased investment in startups by VCs resulted in an all-time high of approximately $4 billion invested in Generative AI startups. Even though the market has tightened heading into 2023, barring OpenAI’s +$10 billion raise, close to half a billion dollars have already been invested in this area. As we progress further into the year, the significant rise in AI adoption by major organizations, which has doubled since 2017, coupled with the noticeable shift from Web3 to Generative AI by both opportunistic founders and FOMO investors, will continue to drive the current market growth.
Key Trends and Drivers
For entrepreneurs seeking to innovate spaces in businesses that leverage sentiment analysis, chatbots, virtual assistants, and similar areas, plentiful and robust foundational models such as DALL-E, GPT-3, and BERT serve as essential building blocks. These models allow for the creation of such applications through simple API calls. Furthermore, the models require significantly less training data than traditional predictive models, adding to their appeal.
Another primary driver behind the advancements of these models is cloud migration, led by major players such as Amazon Web Services, Microsoft Azure, and Google Cloud Platform. The increased computing power provided by these cloud environments was previously unattainable. Additionally, these cloud environments facilitate the adoption of many of these models by both large established enterprises seeking AI adoption and pre-seed startups pushing the limits and exploring the possibilities of these models.
Finally, the year 2023 has already witnessed a significant spending spree by big tech companies on AI. Microsoft has committed $10 billion to advance OpenAI, while Google lost over $120 billion in market value overnight due to Bard’s public blunder. These companies have the financial resources to invest in the research and development of new models which will progressively require more computational power and substantial amounts of proprietary data. However, they cannot achieve these goals alone. Entrepreneurs with an appetite for innovation must take the lead and build daring and exceptional products.
Competitive Landscape
Here are some of the major players in this space:
OpenAI’s GPT model is utilized for natural language processing (NLP), comprehending and generating natural language, including English, French, Spanish, and others. Their DALL-E model can generate realistic images and artwork from text prompts.
Google’s BERT model is used for NLP, enabling computers to understand language much like humans do.
Meta has produced OPT, a set of open-sourced models that perform similarly to GPT.
Microsoft’s Turing NGL model, like GPT, can generate words for completing open-ended textual tasks, document summaries, answering direct questions, and more.
Several of the noteworthy smaller players to keep an eye on include:
Huggin face’s Bloom has the capacity to generate text in 46 natural languages and 13 programming languages.
EleutherAI’s GPT-NeoX-20B is a general-purpose autoregressive language model that forecasts the next word from a context given a set of words.
Challenges within the Space
Generative AI models may be unreliable and generate inaccurate information, as demonstrated by Google’s recent experience. For instance, when prompted about the number of Cloudflare founders, Chat-GPT inaccurately reported that there were only two founders when, in reality, this was not the case. These models may occasionally produce output that does not meet expectations, with no clear explanation, thereby posing challenges in identifying the most appropriate remedy for the issue.
There are a multitude of ethical concerns associated with various generative AI models. One major concern is the potential for these models to create Deepfakes, or synthetic media, which can falsely represent individuals’ likeness through generated images and audio. The use of Deepfakes to spread disinformation has become increasingly prevalent, as evidenced by instances such as the distribution of two fake news anchors by pro-China bot accounts on Facebook and Twitter to stir up tensions surrounding sensitive topics like mass shootings and US international relations, referred to as Wolf News.
Ensuring data privacy is a significant challenge that must be addressed before the widespread adoption of Generative AI models can occur. There are several areas of concern related to data privacy, including data collection, storage, and sharing, lack of transparency, bias, discrimination, and security risks. Many AI models are trained using publicly available data from sources such as social media accounts, public forms, and more. Regulatory bodies have failed in the past to prevent tech companies from becoming too big, making players in the AI space a possible target for future regulations. This could potentially hinder the progress of AI research and development in the future.
What does this all mean?
The global market for AI is anticipated to witness substantial growth in the upcoming years, owing to the growing implementation and development of fundamental models, as well as the presence of $531 billion of unallocated capital in the venture capital asset class as of 2023. Additionally, investors in the technology industry have been impacted by the recent cryptocurrency market crash and are searching for the next prominent investment opportunity. In this regard, banks such as Credit Suisse are offering a 6.5% annual interest rate on new three-month deposits of $5 million or more. As a result, limited partners will likely exert pressure on venture capitalists to invest their capital, or they may look for other investment options in the future.
From my initial experience with AI to working on innovative applications that leverage AI/ML at Lockheed Martin, I have maintained a positive outlook on the field. History suggests that most AI startup investments may fail, similar to the dot-com boom. However, if AI proves to be the next major technology trend similar to mobile and cloud computing, the power law could rescue select VC firms. This means that the next generation of major tech companies such as Google and Facebook could emerge in the near future, producing significant returns and a long-lasting impact.
For more tools and resources ➡️ Visit StartupStash.
Zendesk is giving $75,000 in credits and perks for startups! ➡️ Apply Now!
