Welcome to over 2,000 new subscribers joining us this week. In Today’s Newsletter:
Tech Corner: AI Hype Has Gone Too Far
Growth Corner: Now is the Time to Double Down on Your Flywheel
Paid Corner: How to Build a Flywheel
AutoGPT and the Illusion of Artificial General Intelligence
AI hype is slowly resembling the web3 and metaverse hype trains before it. No matter where you look on social media, you see folks making bold claims about the recent advances in AI.
Three sad realities of growing a social media following converge in the case of trends like metaverse in 2021, web3 in 2022, and AI in 2023:
Writing about trending topics works.
Making readers feel the emotions of wow, hope, and “I can’t believe that!” make things go viral.
Companies building on the latest trend are eager to advertise with large accounts.
This means that content creators have every incentive - from growth to engagement and monetization - to be as excited and uncritical as possible. As a result, every progression in AI is heralded as the next big thing.
The latest dubious claim from this hype train is that AutoGPT signals a turn to Artificial General Intelligence (AGI). This is usually accompanied by ominous pictures of Mr. Smith from The Matrix.
But, the fact is: GPT-4 and AutoGPT aren’t close to artificial general intelligence (AGI).
GPT-4 is dumb. It’s like the person at work who always sounds good, but never actually guides the company toward good decisions. GPT-4 sounds really smart. But it’s just a large language model (LLM). Its faculty with language and code has pulled the veil over everyone’s eyes.
GPT-4 can write beautifully for you. It can reconstruct what you want to say in one-hundred different ways. It can search the web and piece together intelligence from different parts of the web much faster than any human.
But once you ask GPT-4 to make hard decisions for you, it loses any ability to actually analyze the situation. That’s because it’s just a language model. It lacks a logical-semantic underneath to understand what it’s writing.
Auto-GPT is even dumber. For the 99.9% of people who can’t get off the GPT-4 waitlist, it runs on GPT-3.5. So it’s already at a disadvantage. On top of that, it’s just a large language model that can run other large language models. Its ability to give itself tasks has fooled everyone into thinking it’s a breakthrough.
GPT-4 and Auto-GPT have many limitations as artificial intelligence systems. They do not have a deep or broad understanding of the meaning or context of what they generate. They do not have a robust or flexible ability to reason, learn, or adapt beyond their training data. They do not have a clear or reliable way to explain their outputs or correct their mistakes. They do not have a direct or effective way to interact with the real world or achieve any goals.
These are some of the essential features of artificial general intelligence (AGI). AGI is the ability of a machine to perform any intellectual task that a human can do. AGI is not just a bigger or better version of GPT-4 or Auto-GPT. AGI is a fundamentally different kind of intelligence that requires new approaches and paradigms.
So don’t fall for the hype that we have “baby AGI.” Most of that is actually just semantic confusion, because one of the AutoGPT projects is named BabyAGI (made by my inimitable Twitter friend Yohei Inakajima).
These projects are a great step towards AGI, but they are not AGI.
The Era of Integrated Generative
The era we are in now is the era of integrated generative AI. Large language models and image generation models have hit impressive inflection points in the past few months.
This makes them essential features to incorporate into most types of software being built today. However, their value and benefit come from training them on proprietary company-owned data and directing their attention to solving your customer’s specific problems.
There is going to be generative AI helping accountants, helping software engineers, and helping writers. But it’s not going to replace their jobs anytime soon.
Integrated Generative Should be AutoGPT
AutoGPT and BabyAGI are really, as the co-founder of OpenAI said, “the next frontier of prompt engineering.” They offer an implementation on top of GPT that is a great model for any product builder to add to theirs:
The chaining of instructions to itself
The ability to continue operating autonomously forever
These are capabilities that OpenAI should probably just build itself. But until then, they are great ways to integrate generative more intelligently.
Greg Isenberg’s Twitter thread on AutoGPT provides three really controversial example applications: customer service, social media, and financial advice. The thing is, you actually don’t want GPT-3.5 doing tasks as important as these for you. I barely trust GPT-4 doing them. Do you really want an autonomous GPT-3.5 agent doing them for you without review?
Maybe as a small or medium business (SMB), but not in the enterprise. You want a very sculpted and scripted set of instructions around AutoGPT to do your customer service, social media, or provide financial advice. And that’s where the market opportunity for companies and product builders lies.
Take the LLM advances from GPT-4 and AutoGPT, and build them more intelligently into your product, with actual knowledge of the problem at hand.
Products that are doing this well include Microsoft Office, Notion, and Adobe Photoshop. They’ve each integrated the LLMs directly into their core product value. That’s the type of stuff that really works.
Now is the Time to Double Down on Your Flywheel
One of the most interesting topics in product growth that I have been writing about for the past few years is that of the flywheel. This week, I shared the story of Amazon’s flywheel strategy on social media, and you all ate it up.
That’s because, in these tough tech markets, the need to double down on your flywheel is more important than it has been in 20 years. As Jim Collins said in 2001:
In times like this you want to respond not by reacting to bad news, but by building a flywheel.
Tech is having a similar moment to that 2001 period: times are tough. But that also means the companies that grow with maniacal focus could have the trajectory that Amazon has had.
It’s the perfect time to double down on your flywheel. It can enable your exponential, Amazon-like growth curve:
So, the next question naturally is: how do you build a flywheel? And then, how do you double down on it? Here’s exactly what you need to do.
Product Growth is a reader-supported publication. I am working on this full-time. So, this 2,500 word deep dive on how to identify and build your flywheel is for paid subscribers only.
Keep reading with a 7-day free trial
Subscribe to Product Growth to keep reading this post and get 7 days of free access to the full post archives.