A request for my loyal readers: Don’t forget to like and share this post. It greatly helps with visibility across the Substack platform. Thanks, Steve.
On with the show!
We are in the earliest of days of AI’s effect on business transformation. While my older pals talk about working on AI in the 1990s, AI has, until recently, been the realm of technology researchers and engineering specialists. It wasn’t until November 2022 that the investments in this field finally paid off. The layperson could now understand how AI worked by seeing it magically generate text and images.
The release of ChatGPT in November of 2022 was the equivalent of the early 90s for the internet. It wasn’t until the bubble burst in 2000, that we really understood what use cases would truly be transformed by this new technology. It’s 1993 for AI. We finally have AI that is functional in a way that can be understood by non-tech people. Now we have to figure out how to use it and investors are standing in the way.
The Hype is Real
Gartner invented the Hype Cycle as a way of talking about how new technologies are considered. The first two stages of the hype cycle are the Innovation Trigger and the race to the Peak of Inflated Expectations
Innovation Trigger: A potential technology breakthrough kicks things off. Early proof-of-concept stories and media interest trigger significant publicity. Often no usable products exist and commercial viability is unproven.
Peak of Inflated Expectations: Early publicity produces a number of success stories — often accompanied by scores of failures. Some companies take action; many do not.
The introduction of transformer models in 2017 revolutionized tech’s ability to interpret and use language and, perhaps more significantly, unlocked the imaginations of many application developers. Companies like OpenAI, which had been working on similar problems since December 2015, had the right team at the right moment to take advantage of the transformer revolution. June 2017 was the innovation trigger for today’s AI.
With each generation of technology, the adoption curves are steeper and steeper. It took 75 years for landline telephones to have widespread adoption. Cell phones took only a few decades to become the default. For smart phones, it seemed like days. You don’t have to look to hard to find AI all around us. It seems to be cropping up everywhere. But AI’s isn’t the thing. AI is a feature of the thing. It’s why AI investments to date are staring at the abyss.
Goodbye carburetor!
Electric cars changed automobiles in a way that makes the inner workings unrecognizable from their fossil-fueled forbears. The current generation of AI is not that. While this is a subtle point , it’s not AI that will change everything, it’s AI-augmented technologies that will change everything.
I believe that current transformer and diffusion-based technologies are innovations similar to fuel injection in cars. Yes, the blinking cursor in Gemini’s chat interface is very cool. As were the EFI badges and the quicker acceleration in cars during the 1980s. But it was just a modest improvement in performance and efficiency compared to prior generations. It was cool, but it wasn’t all that different from what came before.
You never see an “fuel injection” badge on cars anymore. It’s gone from an innovation—which, honestly, never really mattered to buyers—to just part of the plumbing. That’s where today’s most celebrated technologies will end up. Large language models aren’t the thing. They’re the gadget that will support the thing.
The greatest innovations are yet to come. I know this because of all the Stupid Buttons™️ that are appearing.
Stupid Buttons
To understand how incredibly uninnovative (is that a word?) the tech landscape has gotten, look no further than your favorite app.
It seems like every app now has an AI button and most of those buttons are powered by language models. These buttons perform various functions that all seem very similar:
In Buffer, the AI button will rewrite your post into something totally unrecognizable.
In Evernote, the button clarifies and organizes your notes.
HubSpot’s content assistant helps automate content tasks.
Very similar to each other and fairly dumb. The Buffer rewrites are rarely satisfying. I don’t really understand what Evernote is doing. HubSpot makes a good effort but it’s clearly early days.
Most of the buttons are there because everyone wants AI pixie dust sprinkled on their tech so they can ride LLM coattails to higher valuations. It’s the surest sign that AI is looking into the Trough of Disillusionment.
Evernote is a tech that I use every day. The hardest thing about using a note taking app is not writing the notes, it’s finding the notes. Sure you can use tags and notebooks and other manual organizing principles, but what I’d really like is a better way to store, retrieve and synthesize information.
Evernote sprinkled LLM pixie dust on their app. It helps better organize and cleanup the notes I’m taking. It’s awful. It’s an AI button slapped on to prove they’re AI powered and it doesn’t do that particularly well.
This is a feature exactly nobody wants. What people like myself want is better search over the corpus. We want connections made between similar and dissimilar topics. We want help with “Aha” moments.
“Hey Evernote, find the notes I wrote when working with EdTech startups last year and summarize those in response to this proposal”
That would be awesome. Not only is organizing my rambling a stupid use of a very powerful technology, it solves a problem nobody has.
18 Months
The past 18 months in the AI space have been all been focused on the technology and demonstrations of that technology. That’s fantastic. I love Gemini, ChatGPT, and Midjourney. But I believe these apps are no more than demos of the power of these technologies. They’re not the product itself.
The next 18 months need to be about product. By turning our attention from the technology to the business problem we can start to unlock the promise of the technology investments in AI. By focusing on the problem space instead of the technology, AI can be truly transformative.
The only fly in this ointment will be whether the venture community can keep up. Tens of billions of venture funding has flowed into platforms, LLMs, and domain-specific foundation models. While these foundational building blocks are necessary, who will build on them? Large corporations will take up a lot of the capacity but create little value.
In order for a thousand flowers to bloom with the help of AI, the venture community is going to have to get out of its collective funk and return to norms. Now that core investments have been made, seed and early-stage companies have to be unleashed on a variety of problems to create 10-100X returns on those core investments.
And yet, when you look at the investor data, it remains a fraction of what it was in the past. Was 2021 a bit exuberant? Sure. But is the current state of investment adequate? No. The little venture that is going out to “startups” is usually going out in large chunks to established portfolio companies or established founders.
I just read a report that showed Seed investing, the sector of the startup ecosystem previously least affected by the investing downturn, is down 23% from a year ago. While overall early stage investing is up slightly, investors are starving the pipeline that will take advantage of all the big AI gadgets.
The only way that investors realize returns on the billions showered on LLMs is to get those technologies embedded in an app ecosystem that drives value for CEOs, CIOs, and shareholders.
Yes, we’re staring down into the Trough of Disillusionment on many AI investments. But those investments weren’t in the problems that matter. The core platform space will rationalize during the coming months (see stability.ai). That rationalization is the trigger for a new set of hype curves. This time, not concentrated in a single, horizontal, but now in multiple verticals.
That’s what I and many other entrepreneurs are focused on. It’s what investors should focus on as well. We don’t need any more stupid buttons. We also don’t need stupid money flowing into core technologies. The past was about the pure science of AI. The future is applied AI.
Next week, I’ll continue the discussion about Google that I started just after ChatGPT was launched. In those early days, I expected Google’s heyday was ending. I still do, but Google’s putting up a good fight.