Beyond the Buzz: A Deep Dive into the Generative AI Horizon

Introduction

On November 30th, 2022, the world witnessed a seismic shift in the technology landscape when ChatGPT, a large language model (LLM), was released, and it took the world by storm. It could engage in conversations that was indistinguishable from those with real people. Yet, amidst the enthusiasm, LLMs and generative AI pose an essential inquiry: Is it merely a fleeting fascination and a tool for trivial amusements like composing poems? or does it hold the transformative power to revolutionize numerous industries and profoundly touch the lives of people worldwide, similar to the way the internet did? Summarizing texts and creating avatars seem fun, but what significance do these capabilities truly hold?

To answer this question and guide our vision of the future, it’s important to ask the following questions:

• What have been the driving forces in reaching this juncture?
• Why is now the optimal time to consider investing and building in this domain?
• What concrete use cases can emerge and which sectors are poised for disruption?
• Where can we identify compelling investment prospects

What are LLMs, and what is their significance

LLMs leverage the transformative power of the transformer model, developed in 2017 by Google researchers. This technology breaks down word sequences into tokens, enabling the model to understand the context by encoding and training on extensive datasets from internet text. The unique self-attention mechanism within the transformer architecture distinguishes LLMs and allows it to capture context comprehensively and understand relationships between words more effectively compared to previous models like recurrent neural networks (RNNs).

Recognizing the nature of LLMs reveals that they function more as pattern-spotting engines than fact-checking systems. They excel at generating text that appears plausible but may not always be factually accurate. However, given the vast number of parameters on which these models are trained on and the map of the world they have created in text, they represent the closest approximation to mimicking the human brain’s capacity to discern patterns and comprehend context. Presently, they surpass human performance in speech and text recognition, reading comprehension, and language understanding.

The proficiency of LLMs in understanding natural language, their capacity for reasoning, and their understanding of context, has surpassed expectations and marks a new era in human-computer interaction.

Just as the graphical user interface transformed personal computing in the 1980s, LLMs with natural language interfaces will democratize access to AI and redefine how we work and interact with software.

Why now

It’s important to study past cycles of innovation to discern patterns and filter through the noise. Historically, we have witnessed that major economic shifts and market dislocations occur when there is an irresistible economic incentive (either in revenue expansion or cost reduction) or when new enabling technology improves product experiences by a factor of 10X. The fact that we haven’t yet witnessed a platform shift similar to the mobile and internet revolutions in AI suggests that the use cases have been somewhat limited in scope, not yet serving as platform enablers for disrupting entire sectors.

However, we are standing at an inflection point where the economics of generative AI are incredibly compelling. Looking back at the previous era, we can draw parallels to the current situation. The microchip era dramatically reduced the marginal cost of computing, reshaping iconic companies like IBM. The internet similarly drove distribution costs to near zero, giving rise to tech giants like Amazon and Google. More recently, the mobile era decreased the cost of access and connectivity, allowing users to connect at zero marginal cost, giving birth to companies like Facebook with new business models. And now AI and large language models have dramatically reduced the marginal costs of production, making us stand on the brink of another transformative moment, one that will result in massive productivity gain.

If we look at the average number of workers at an S&P 500 company needed to generate $1M, revenue has decreased by 75%; adjusting for inflation, this number could drastically improve over the next decade as individuals and businesses do a lot more with fewer resources.

As we navigate this inflection point, the excitement lies in the promise of generative AI evolving from novelty applications to truly transformative products, with the potential to disrupt industries on a broad scale. According to Goldman Sachs, Global GDP could grow by an extra 7% or $7 trillion and lift productivity growth by 1.5% over a 10-year period.

The value chain of Generative AI

To grasp the evolving landscape of the generative AI market, it’s crucial to delineate the current structure of its technological framework. It can be dissected into three fundamental layers:

The application layer

The application layer could be broken down into two primary categories:

1. Horizontal Tooling: Aims to dominate workflows or other processes across various industries. Examples include Tome, which provides AI-driven presentation solutions.
2. Vertical Turnkey Applications: Tailored for particular verticals or sectors, delivering comprehensive solutions within one specific industry. A notable example is Cradle, which offers AI-assisted tools for protein design.

Why the Application Layer Is the Most Exciting Area

While the infrastructure and model layers will continue to play a very important role in shaping the future of Generative AI, our greatest enthusiasm lies in the application layer with its potential to reach a wide audience in various areas.

The evolution of applications has unfolded in two distinctive waves. The first wave was characterized by initial hype and excitement, akin to a “me too” phenomenon, where everyone sought to partake in the burgeoning field. During this phase, numerous niche applications and autocomplete features emerged with no distinct competitive advantage. However, as:

1. The quantity and quality of parameters these models are trained on increase,
2. Models become more complex with user feedback as well as new techniques such as ‘reflexion’ and ‘grounding’ enhance AI’s capacity to provide more accurate predictions and responses using specific, contextually relevant information,
3. Efficiency is improved in managing computing and storage costs,

the divide between customer expectations and Ai’s capabilities diminishes, facilitating domain-specific use cases exemplified by innovations like Harvey and Inceptive.

This leads to the second wave, promising to deliver deep value and enriched experiences to end customers by using foundation models as part of a more comprehensive solution rather than the entire solution. This phase represents a more mature and refined stage of generative AI, where the focus shifts from novelty to the substantial enhancement of user experiences and real-world applications.

During the initial wave, those who triumphed were the ones utilizing general-purpose LLMs for use cases where precision wasn’t crucial. However, as models progress, we anticipate the rise of novel applications demanding greater precision and complexity, thereby unlocking fresh opportunities in the market.

Five Focus Areas within the App Layer:

1. Revolutionizing Healthcare

The U.S. spends a staggering $4 trillion annually on healthcare, with a substantial $1 trillion directed towards administrative costs and the remaining $3 trillion allocated to care delivery. Large Language Models (LLMs) present a transformative solution for reducing healthcare administrative expenses by automating tasks such as report generation, data summarization, and processing insurance claims. LLMs also contribute to patient education, enhance telehealth services, and improve workflow efficiency, thereby freeing up time for healthcare professionals to focus on delivering better patient care.

Implemented by companies such as Bayer Pharma, HCA Healthcare, and MEDITECH, these models showcase their ability to expedite clinical trials, improve documentation accuracy, and speed up processes, thereby demonstrating their potential to streamline administrative procedures and drive significant cost reductions in the healthcare sector.

Beyond administrative tasks, we’ll see clinical agents capable of conducting initial screenings and suggesting diagnoses to doctors. By efficiently managing administrative duties and performing basic screening, these models can free up valuable time for healthcare professionals. This, in turn, enables them to concentrate more on patient care and complex medical tasks, leading to increased productivity and improved care delivery. Other AI assistants for healthcare workflow automation, such as Navina and Synthpop.ai, also demonstrate how administrative tasks can be effectively automated.

1. Transforming software interaction

The current efficiency of human interaction with software remains suboptimal. We still have to navigate through multiple apps to plan for a vacation or book an appointment. However, LLMs presents an opportunity to redefine our daily interactions with software. Think of interactive AI, a new paradigm where AI systems can act on high-level goals, interact with other AIs, and potentially collaborate with humans to achieve tasks. This shift has the potential to reshape how we interact with technology and automate various processes, offering exciting investment opportunities. Interactive AI is capable of carrying out tasks beyond text-based conversations. These AI systems will have the ability to execute tasks set by users, utilizing other software and even involving human collaboration. This shift marks a profound transformation in technology, where AI will have a level of autonomy to take actions on behalf of users. It opens new possibilities for more dynamic and efficient interactions with technology. Companies like Inflection.ai and Hu.ma.ne are pioneering this interactive AI wave, and they represent an exciting frontier for investment and innovation.

1. Elevating Digital Production

Generative AI extends beyond LLMs and text, encompassing diverse models like diffusion models, that are text-to-image creators, enabling the creation of images, videos, and audio. These are valuable assets for professionals involved in tasks such as game production, marketing illustrations, video creation/editing, interior design etc. Additionally, this transformative era will enable individuals to stand out for their ideas and creativity, breaking free from traditional skill constraints, echoing the way Canva once democratized graphic design for non-artists. The democratization of creative tools will empower everyone to tell their own stories, fostering a more engaging and personalized digital world. A good illustration of this is Runway ML, bringing a production studio to consumers’ fingertips. Other notable examples include Genmo for Short-form video, Soundraw.io for Music, Sloyd in 3D asset generation.

1. Boosting productivity

Utilizing LLMs and generative AI has the potential to transform personal efficiency by tackling daily challenges, ranging from booking a flight and a hotel, to recommending new destinations based on your life circumstances, and placing an order from Doordash without needing to open the app. The prospect of AI-powered personal assistants or coaches holds vast potential, envisioning an AI-driven companion capable of optimizing time and resources. One where a user can have a much more nuanced conversation. These AI advancements promise to significantly improve efficiency, decision-making, and workflow streamlining, impacting both personal and professional domains. From assisting in entertainment recommendations to automating tasks like sending emails and schedule management, helping you choose your next TV and many more everyday use cases to help you spend more time on creative tasks. Notable examples include Microsoft Co pilot for work and Ninja.ai where their AI assistants aim to efficiently support professionals by acting as their executive assistant.

1. Democratizing Education

As we examine the landscape of student needs, there are four crucial problem areas: Access, Content, Assessment, and Personalization. While companies are actively addressing the first two aspects, the needs in Assessment and Personalization remain underserved, presenting untapped opportunities for innovation. By unlocking the skills data needed to drive success and by a contextual understanding of a student’s needs to deliver personalized learning experiences through tailored content, adaptive learning paths, and automated grading, generative AI has the potential to redefine education.

The promises made by CEOs of leading companies such as Coursera, edX, Pearson, and Anthology reinforce the transformative impact anticipated in the Ed-tech space. Anthology’s collaboration with OpenAI to develop tools like a course-building aide and Pearson’s commitment to enhancing the Pearson+ app with ChatGPT-powered features, including automatic summarization of video content and AI-powered chatbots, showcase the tangible applications of generative AI in addressing educational challenges. Companies like edX and Coursera, with their ChatGPT plugins, underscore the industry’s commitment to providing learners with personalized assistance, learning materials, and career guidance through the integration of generative AI.

Conclusion

As we navigate this transformative phase, we’re left with more questions than answers. What will define the new competitive advantages, and can AI-native startups be able to win major segments of the market in this evolving landscape or will incumbents leverage their distribution network advantages to dominate the market? With increased adoption, who will own customers’ data and how do we address privacy concerns?

Despite these uncertainties, the transformative potential of generative AI in revolutionizing creativity, productivity, and software interaction is evident. The future looms where every individual will have AI companions amplifying their capabilities and augmenting their intelligence. This integration of AI not only foresees heightened productivity and economic growth but also a renaissance in creativity and cost reduction. The trajectory from text-based conversational interfaces to emerging modalities like generative user interfaces and human-sounding voices signifies a paradigm shift in how we interact with digital information.

The strategic approach to investment in this realm demands a nuanced understanding of the technology’s short-term and long-term impacts. The looming challenge lies in striking a balance between recognizing the transformative power of generative AI and navigating the risks associated with its implementation, with an eye on where economic moats will be deepest and value pools will accrue in the years to come.

Our investment strategy is rooted in the belief that AI advancements will create new opportunities and disrupt existing sectors while forming new ones. However, we’re mindful of the patterns seen in technological revolutions, characterized by cycles of excitement, frenzied investment, asset bubbles, and market corrections. As we venture into this dynamic era, we remain vigilant, adaptable, and committed to harnessing generative AI’s potential while navigating its challenges with caution and foresight.

Thank you for reading. Would love to hear your thoughts and feedback. In this evolving landscape, uncertainties abound regarding the future and which companies will emerge as true winners. The only sure way to unravel these mysteries is by embarking on the journey of building and experimenting. If you share a passion for navigating uncharted territories, I invite you to connect.