The Rise of AI Collaborators

Welcome Shareholders,

Every Sunday, We'll highlight one company to do an in-depth breakdown of their business, financial situation, and whether or not it's a worthy investment opportunity - large or small. In today's edition, we'll look at a startup looking to usher in a new age of productivity by building a Universal AI assistant to help knowledge workers automate most routine tasks.

Over the past decade, several concurrent technology trends have reshaped the world and the way we live and work. Smartphones, cloud computing, and the internet itself have fundamentally transformed our lives, with far-reaching impacts in various industries. However, the introduction and popularization of artificial intelligence (AI) is poised to create a tectonic shift, especially for knowledge workers. This shift is expected to eclipse the scale and speed of changes that were brought about by previous technological revolutions.

While exciting AI developments like OpenAI's text generation platform ChatGPT and text-to-image generators like Dall-E 2 and Midjourney have captivated the world in recent months, they represent just the tip of the iceberg. While generating artwork and writing are great applications, the real game-changing moment will come when AI transcends such novelty applications.

The future of AI lies in its ability to empower enterprises and knowledge workers by automating mundane tasks and unlocking numerous levers for improving productivity. When AI enables knowledge workers to seamlessly execute any task they can articulate, it will effectively democratize expertise and capabilities once reserved for a select few.

Enter Adept AI, a startup focused on harnessing the full potential of AI to revolutionize productivity for enterprises and knowledge workers alike. Today we'll look at how Adept AI is looking to pair humans and machines seamlessly together to bring a new era of efficiency and innovation.

Unleashing The AI Genie

A decade ago, the idea of a universal AI assistant seemed like a distant dream. Fast forward to today, and Adept AI has emerged as a promising startup on a mission to make this dream a reality. In April 2022, Adept AI came out of stealth mode, raising $65 million in funding with a focus on building general intelligence that enables humans and computers to work together creatively to solve problems.

Adept AI was founded by former Google employees who played a crucial role in developing the popular transformer architecture. The team, led by co-founder and CEO David Luan, aims to teach machines to use every software tool and API in the world. CTO Niki Parmar and chief scientist Ashish Vaswani, along with an early group of employees, left their previous roles at Google and DeepMind to join the Adept AI team. The co-founders boast a deep understanding of transformer-based systems. Luan was involved in the development of GPT-2 and GPT-3 at OpenAI before joining Google, where he worked with Parmar and Vaswani. The pair is credited with developing the AI transformer architecture, first proposed in 2017.

At Google, Luan served as the tech lead for the "large models effort" at Google Brain, one of the tech giant's preeminent research divisions. Luan's team aimed to build one general model to power all machine learning use cases by training larger and larger Transformers. However, they encountered a significant limitation: the best results were domain-specific and tailored for tasks like analyzing medical records or responding to questions on particular topics.

According to Luan, the transformer was the first neural network that appeared to work for every major AI use case, which convinced him that general intelligence was possible. The team trained increasingly larger transformers with the dream of creating one general model to power all machine learning use cases. However, they faced a clear limitation: models trained on text could write great prose but couldn't take action in the digital world. To overcome this limitation, Adept AI will continue to build on the current transformer architecture. Luan explained that, unlike text-only models, they are training their AI on actions taken on a computer. The tasks and tools Adept AI will control will start simple and become increasingly complex over time. Luan envisions a future where developers could leverage the model to brainstorm and even carry out scientific research.

What is Adept Building?

Adept AI's co-founders envision a future where their AI system serves as an "overlay" within computers, utilizing the same tools that people use daily. This AI overlay would respond to commands like "create a marketing strategy presentation" or "analyze sales data for Q3 trends," all while using existing software such as PowerPoint, Excel, Salesforce, and Google Analytics to complete tasks.

The core of Adept's ambition lies in training a neural network to use every software tool in the world, building on the vast array of existing capabilities that people have already developed.

The idea is that with Adept AI, users can focus on the work they enjoy most, delegating other tasks to the AI system. Adept AI's collaborator is expected to be a good student and highly coachable, becoming more helpful and aligned with human interaction over time.

At first glance, Adept's creation might seem similar to robotic process automation (RPA)—software robots combining Automation, computer vision, and machine learning to automate repetitive tasks like filing forms and responding to emails. However, the Adept AI team maintains that their technology goes far beyond what RPA vendors like Automation Anywhere and UiPath offer, providing a more sophisticated and integrated solution for empowering knowledge workers and enhancing productivity.

The Pitfalls of RPA

While the concept of workforce augmentation in the enterprise has been around for some time, previous technologies, such as robotic process automation (RPA), have not quite lived up to expectations. RPA has faced several challenges that hindered its ability to deliver on its full potential:

1. Complex implementation and insufficient documentation have led to maintenance nightmares.

2. Frequent changes in underlying applications create instability in the RPA ecosystem.

3. Resistance from business users and sponsors, who may be skeptical about RPA's benefits.

4. Hype-driven and unplanned deployments, often without proper testing, have resulted in suboptimal outcomes and continued dependence on human workers.

These pitfalls raise the question of whether Adept AI can succeed where RPA solutions have fallen short. If Adept AI can overcome these challenges, the potential rewards could be enormous. As per a recent industry report, the business process automation technologies market, which streamlines enterprise customer-facing and back-office workloads, is expected to grow at a compound annual growth rate of 11.8% over the next decade, from $13.7 billion in 2023 to $41.8 billion.

There are other tailwinds pushing process automation forward, including the increased prevalence of remote work, pushing a majority of firms to increase investments. Adept AI can capitalize on this growing demand if it can address the limitations of traditional RPA and deliver a more robust, adaptable, and user-friendly solution.

True General Intelligence

Adept AI is on a mission to break free from the limitations of RPA and forge an AI system that serves as a true collaborator for knowledge workers. With the ambitious goal of achieving true general intelligence, Adept aims to empower users to fully harness the power of AI, unlocking unprecedented levels of productivity.

Picture an AI teammate that goes beyond merely reading and writing text, venturing into the realm of taking action in the digital world. Adept AI is striving for this vision, aspiring to create a universal AI collaborator that can effectively engage with a plethora of software tools and APIs, all while responding to users' needs and desires.

To turn this vision into reality, the Adept team plans to employ reinforcement learning techniques to train their models in the use of standard software tools. This approach distinguishes Adept AI from other startups concentrating solely on text or image generation. By focusing on training an AI model to comprehend how humans use computers and navigate the software, Adept AI lays the foundation for a system that can transform text instructions into actionable digital tasks, thus bringing AI a step closer to true general intelligence.

Aware of the potential pitfalls of AI, Adept's team is committed to selecting training data that emphasizes the software tools the model is designed for, thereby reducing the propagation of human biases that have haunted large-scale machine learning models.

MVP and Path to Commercialisation

Adept AI's Action Transformer is designed to take actions on behalf of users, providing a powerful solution for those whose workdays are filled with repetitive, low-yield tasks. The ultimate goal is not to replace knowledge workers but to empower them by placing them in the driver's seat with an AI assistant, enhancing decision-making, sparking new solutions, and freeing time for more fulfilling tasks.

Adept's Minimum Viable Product (MVP), ACT-1, can perform tasks like importing LinkedIn URLs into recruiting software, displaying as an overlay window on top of existing applications such as Google Chrome or Salesforce.

The prototype is currently available for desktops, with a mobile version expected shortly. The versatility of ACT-1 has attracted strategic investors like Microsoft, Nvidia, Atlassian, and Workday, all of which market software that could benefit from this AI assistant.

By potentially lowering the barrier to entry within the enterprise workforce, ACT-1 may contribute to greater inclusive prosperity. While the public launch of the AI assistant is yet to be announced, Adept AI already has revenue commitments and targets from a handful of partners.

Competitive Landscape

Adept AI's pursuit of AI-driven collaboration between humans and computers is not without its fair share of competition. DeepMind, backed by Alphabet, has already made waves in the industry with its AI research focused on "instruction-following" tasks. In a paper published in February 2022, DeepMind scientists showcased an AI system that learned how to perform tasks, such as booking a flight, by observing users' keyboard and mouse commands.

Joining the fray is Inflection AI, a collaboration between DeepMind co-founder Mustafa Suleyman and LinkedIn co-founder Reid Hoffman. Inflection AI shares Adept AI's ambition to harness artificial intelligence for enhancing human-computer collaboration and improving work efficiency. Adept AI could also face challenges from its co-founders Ashish Vaswani and Niki Parmar, who recently departed to another startup. This turnover may affect the company's progress and ability to stay competitive in the market.

AI's App Store Moment

Perhaps the most significant competitive challenge for Adept could come from OpenAI in the near future. Last week, OpenAI's ChatGPT took a significant leap in capabilities by introducing third-party plugins. This transformative moment for OpenAI echoes the iPhone App Store's launch, revolutionizing app development and distribution. Similarly, ChatGPT's plugin ecosystem invites developers to design and share extensions, enhancing user experiences and unlocking a myriad of new possibilities for AI-powered chatbots.

A standout plugin is the web browser plugin, which enables ChatGPT to access real-time data from the web, overcoming its previous knowledge limitations. This plugin allows ChatGPT to provide accurate, up-to-date information by citing the sources it visited while crafting answers. Other partners include Expedia, Instacart, Kayak, OpenTable, and Zapier.

For example, the Expedia plugin enables users to design travel itineraries through a chat with ChatGPT, utilizing real-time availability and pricing data from Expedia. Another exciting plugin comes from Zapier, which connects ChatGPT to various productivity tools like Google Sheets, Trello, and Gmail, enabling users to trigger various tasks.

The introduction of third-party plugins for ChatGPT empowers users to create personalized AI assistants that become increasingly intelligent over time. This is a significant advantage for OpenAI, given the importance of user data in enhancing large language models. As a result, most companies may prefer partnering with OpenAI due to its extensive user base, potentially hurting Adept's enterprise ambitions. The introduction of plugins in ChatGPT is a major curveball for Adept, which would have previously assumed that its capabilities and commercial applications primarily targeted a different market than that of OpenAI. However, as it's becoming increasingly clear, OpenAI wants to rapidly expand its capabilities and quickly target all kinds of AI customers.

Looking Ahead

Adept's CEO, David Luan, shares the ambitious goal of achieving AGI with his former boss, Sam Altman of OpenAI. However, Luan envisions a more business-focused AGI with humans remaining in the driver's seat. This approach reflects Adept's dedication to empowering knowledge workers and enhancing productivity without diminishing the importance of human input.

To support its vision, Adept recently raised $350 million in a Series B funding round in March 2023, co-led by General Catalyst and Spark Capital, with participation from prominent investors. This funding round reportedly valued the company at "at least" $1 billion, bringing Adept's total raised to $415 million. The capital will be used for product development, model training, and expanding the company's headcount.

Adept seeks to build on the momentum generated by giant foundation models for language and images in recent years. Its approach involves creating a new kind of foundation model capable of performing actions on any software tool using natural language. By doing so, Adept aims to maintain its relevance in the rapidly evolving AI landscape and continue its pursuit of true general intelligence, despite the growing competition from OpenAI and other industry players.

Bottom Line

As the AI landscape evolves at a rapid pace, Adept AI's pursuit of true general intelligence has the potential to redefine the way knowledge workers interact with technology. By focusing on creating a universal AI teammate, Adept AI could pave the way for enhanced productivity, better decision-making, and more efficient workflows across various industries. A few factors will determine its long-term success, including the company's ability to adapt and innovate in the face of new developments in the broader AI space.