Peak Hype = Convergence of broad potential applications + significant funding + media excitement + scientific breakthroughs + commercial interest
Well that was fast.
“I guess genAI really is a bust.”
“I told you so!”
Well of course! Just look at that equation. There’s that word again.
Broad potential applications
The dreaded potential.
Potential- what my old trading bosses used to say was the french word for “hasn’t done shit yet.”
However, something to remember.
It’s really, really early in this most recent round of bright shiny object AI.
Remember all the money that went into early ARPANET in the ‘60s? No, me neither.
Remember all the early government money that went into that thing in the ‘70s called Global Positioning System? Nope. Doesn’t ring a bell.
How about that big, cool campus in NJ where they did a bunch of research and development on things like semiconductors? What was the name? Oh yeah, Bell Labs.
The Internet, GPS, and the chip sector.
Lot’s of up front spend, peaks and disillusionment along the way, and they all turned out to be pretty important. But it took some time. And they were also able to build, research, rinse and repeat largely out of the public eye, certainly not to the extent that they would be now.
GenAI is subject to a ridiculous amount of hype.
Hype is a smaller part “what’s actually happening now?”, and large part pulling future expectations forward.
There’s been a remarkable amount of innovation and growth in the space since GPT came into the zeitgeist a little of a year ago.
A year.
“ChatGPT is mildly embarrassing at best. GPT-4 is the dumbest model any of you will ever have to use again, by a lot.” -Sam Altman, CEO Open AI, April 25, 2024
1 year.
And GenAI is a bust?
This overestimation is actually a “thing” called Amara's Law, named after researcher and futurist Roy Amara.
"We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run."
GenAI is real and only going to get better. Why?
Models will get larger and more complex, but at the same time people will work to make them smaller and more explainable.
We will(are) run(ning) out of data to train them. Some people will worry about recursive model collapse, but meanwhile other people will come up with ways to train them with augmented, synthetic, or some other kind of data we haven’t thought of yet.
Token generation is much more expensive than a search engine query, but the cost will come down.
We need lower rates of hallucinations. This is coming.
Current architectures will reach their utilization capacity and limits, but researchers will dust off new-ish ways (but not necessarily GPU friendly) and find brand new ways of approach.
A myriad of things no one has thought of yet are going to be built, are going to happen.
Llms represent the embryonic state of what is to come.
In other words, to steal from Jeff Goldblum and screenwriter David Koepp, “(Innovation and GenAI), uh find a way.”
What about some of the hype trains of old that didn’t pan out(yet)?
GenAI isn’t blockchain. GenAI isn’t nanotech. GenAI isn’t NFTs
The world’s biggest excel sheet.
Build cool things, but make them very small! (Apologies to Richard Feynman)
Very expensive art, but as a gif!
I never really understood the potential applications of blockchain beyond a private ledger and I never had enough money to shrink things or buy expensive jpegs, but I’m already using GenAI applications, multiple times a day, everyday.
I don’t know any other way to articulate it than to say we are building things right now. It exists. It is tangible, and it’s only getting better with more use cases and applications.
I hope that competitors dismiss genAI as a hype technology and stick to “the way it’s always been”. I thank you for your alpha!
GPT-5 is rumored to be in the safety testing stages, so (rumors!) another 3-6 months. Will that bring the first real robust agentic AI implementation? Hype is sky high for this one and will help determine whether we bump to another hype peak or slide back down the trough of disillusionment.
Either way.
Ignore the hype train.
Another one will be by any minute.
Build cool things.
Don’t slow down.