A Reality Check on Generative AI
Generative AI has been a buzz word over the last few months and there is unprecedented excitement in the tech-industry on the potential of this new technology to transform how we operate. In this article, I would like to put this technology in the overall context of human evolution and provide a reality check on how the adoption is likely to grow over the next few years.
In one of the recent essays on Stratecherry, Ben Thompson introduced the excellent concept of Idea Propagation Value Chain
The framework explains how ideas spread:
Creation – You have an idea.
Substantiation – You share your idea in form of writing, visual imagery, or oral communication.
Duplication – The idea gets duplicated for example in form of multiple written, audio copies etc.
Distribution – The idea gets distributed in form of books, webpages etc.
Consumption – People discover the idea, process it and share it further.
The evolution of human consumption of ideas has entailed removal of bottlenecks across this value chain in the following sequence:
When humans did not know how to write, they could only communicate orally, writing removed the bottleneck around broader consumption.
Printing press removed the bottleneck around duplication by making it economical to duplicate the idea at scale.
Internet removed the bottleneck around distribution by making it free and easy to share ideas broadly. A newspaper for example only had a limited geographic reach.
The last bottleneck in the value chain around creating and substantiating the idea has now been removed by Generative AI. The biggest bottleneck on ideas has been human creativity and bandwidth which can now be done via Gen AI partially or fully.
While the unbundling of this value chain has happened slowly over hundreds of years, the Generative AI wave is built on the foundations led over the last 60 years in form of compute horsepower (to process data), internet (to build training data) and evolution of cloud computing/mobile phones (to help consume the outputs at scale).
With a good understanding of how Gen AI fits in the evolution of human communication, lets look at the evolving industry landscape today.
There are massive investments across the tech-stack for Generative AI which includes:
Foundational Models – e.g. GPT, Stable Diffusion, Llama2 etc.
Compute and Inference – e.g., Google Cloud, Azure, AWS
Model tuning – e.g., Hugging Face, Amazon Sage Maker
Developer Tools/Infra – LangChain
Apps and Workflows – e.g., Lensa, Character.AI etc.
Given the investments across the tech-stack, the biggest beneficiary has been Nvidia whose stock price has increased significantly owing to the demand for GPU capacity to feed the compute hungry Generative AI Models
While it is undisputable that Generative AI will unleash a wave of creativity and new applications, there is also irrational exuberance in terms of money chasing a new technology. There is no dearth of use-cases owing to the novelty factor, proof of value is emerging at a relatively slower pace. The following charts tell the story.
Starting from left, the first chart illustrates how ChatGPT was the fastest product to get to 100mn users within a period of 2 months. This can primarily be attributed to the novelty factor. The second and third chart illustrate one month retention and daily active users/monthly active users (which are key metrics of product stickiness) – as is evident the stickiness is low for some of the generative AI based offerings. This clearly indicates that as of today, customers have the use-cases to try these solutions, but they do not necessarily solve for their day-to-day problems.
So, what is the way forward for Generative AI offerings. Below are some thoughts:
Generative AI solutions integrated with the day-to-day workflows of users will have more stickiness than stand-alone solutions – Microsoft, Google, and Adobe are moving in this direction and integrating the Gen AI solution in the products customers use daily.
A corollary to the above point is that distribution matters more than technological superiority of stand-alone companies. Hence, big players such as Microsoft, Google, FB, Amazon, Adobe have advantage over start-ups.
Use-cases which provide value to customers will eventually gain traction than use-cases based on novelty factor alone. This means vertical specific use-cases such as Healthcare, Legal and enterprise horizontal use-cases such as Sales enablement, Customer support, Software engineering etc. will get much more traction than toys to play with type of solutions.
Thee is no doubt that Generative AI is a game changing technology which will have an impact equivalent, to internet, mobile phones, and cloud computing. As it has been proven historically, the winning companies of this wave as well will be the ones who solve end-to-end problems for customers supported by strong distribution than the ones banking only on technology.