- Beyond The Dashboard
- Posts
- The Unseen Decline of AI Titans: Is the Bubble Bursting?
The Unseen Decline of AI Titans: Is the Bubble Bursting?
Peeling Back the Hype: What Recent Data Reveals About the Future of ChatGPT and Generative AI Technology
Matt Caramenico
September 05, 2023
Round #3
Think back to December ‘22, and ChatGPT’s drinking rocket fuel and Users seem to be in never-ending supply.
They blasted through the 1m user mark within five days, annihilating Instagram's 2.5-month record climb to 1m users.
Pay no attention to Threads… no one else has.
The press was insanely positive.
People were hooked. Your grandma was going to automate Christmas cards, and the world as we knew it was in freefall.
Bigger than frozen TV Dinners or the microwave oven… this was transformative tech, like Benz with the car in 1888.
Gold Rush ‘23
Once people heard about the gold in the Klondike, it exploded in popularity. 100,000 people migrated into the Yukon over a span of 3 years.
People found gold amounting to a cumulative $1b, based on more recent valuations.
If you wanted something new… The Klondike offered a get-rich-quick scenario for the bold.
Similarly, with the democratization of OpenAI’s GPT3, GPT3.5, and GPT4 modules in an easy-to-use platform…
EVERYONE was now capable of inhumane feats. Content creation at scale, Python scripting w/o prior knowledge, NL generation, mass communication via DMs, building a DnD world when creativity runs dry, etc.
An astounding number of companies are building on this underlying tech or leveraging it in some way.
OpenAI just announced that “80% of Fortune 500 companies had used ChatGPT since its November rollout.”
The hype was intoxicating.
The hype was Meteoric, to say the least, but there’s more to the story not being talked about within the churn and burn editorials - Like -
For anyone paying attention… or for those keeping score at home, we know a lot of this posturing is just classic smoke and mirrors.
Every VC with billions sunk into generative AI, either with OpenAI or a competitor, needs the space to go full bore “Social Media” style adoption.
Thing is… It’s not.
Early adopters and Power Users like myself - bumped into the guardrails, and the facade broke a long time ago. Read parts 1 & 2 of this ongoing segment.
What’s the story:
The data tells a different story - alluded to by these -
The sacred J curve broke. Users are declining Month Over Month.
OpenAI is losing its user base…
And considering its status as one of the most democratized/easy-to-use platforms for mass interaction, it's hard to believe in the AI-driven future we all imagined a few months ago.
Sparktoro - Just dug into a 3rd party data set based on millions of anonymized users and published a closer look.
Since May, OpenAI traffic has declined 29.15%
In this deep dive, they hypothesize about the “why” behind this decline:
Theory A) Professional/regular use is still at an all-time high or growing; it’s only new users/those-checking-it-out-with-no-specific-tasks that are falling off
Theory B) A huge amount of use is connected to schoolwork, and educational users in the US, Canada, and other countries with summer vacations are responsible for the decline mirage
Ultimately landing on neither.
I, however, think we can pull back up this study - “How Is ChatGPT’s Behavior Changing over Time?” Published a month ago on August 1st ‘23.
Chen, Zaharia, and Zou over at Stanford and UC Berkeley put in the work so we don’t have to.
The analysis is clear. ChatGPT and its two most popular modules are rapidly changing, and NOT for the better.
Module Accuracy/Time
Feel free to read through the paper at length.
It’s insanely interesting and worthwhile for those interested in the details.
Suffice it to say.
We’ve got these LLMs talking to Bozos at scale, and due to the frequency of bad info & bozo convos, we’re circularly training LLMs with that interaction, and the result is a blade that’s losing its edge rapidly.
Thus, I think fewer and fewer people are using it for meaningful tasks… and 10 months in… the novelty has worn off for the masses.
Where do we go from here?
For me, it’s simple.
This tool (LLMs as a Toolset, not even ChatGPT) was never going to be the world-shaking technological shift people claimed it would be.
This is the 3rd iteration, and my perspective holds true.
A month and a half ago, I said:
“Ultimately, if the current Google SGE product sticks, I think we’ll see a low-impact version deployed that shifts the presentation layer more than anything - less impact on underlying data or how pages “rank” - for now.”
It’s time to update since Google announced deployment and changes to SGE.
SGE should complement traditional search and not replace it.
They will increase the number of links within SGE's response to primary sources.
They are also rolling out SGE to Japan and India.
To me, this falls exactly in line with what I’ve predicted.
The cliff that some within the SEO industry claimed was quickly approaching will never arrive.
And why should it?
The Frankenstein Monster - LLM buckled into the search engine - has yet to prove its usefulness outside of novelty, and the verdict is still out on Generative response tech and how beneficial it is for enabling users to find what they need while fitting into products people are comfortable using.
Global Search Engine Usage
Until next time - Join me as I build - Beyond the Dashboard and share what's on my mind.
Matt
PS: I am still over here building the content generation habit and finding the proper cadence for getting these shipped weekly. One day, I’ll find something that works.
Cool Cars:
Nineteen hours left on this mint targa. I’m guessing $130,000 on this one, but I wouldn’t be surprised if it sold for more.
Cool Links:
New Belgium hosts the Tour de Fat every year. What a fun event up in Fort Collins, Co. Highly recommend getting out there and enjoying a nice bike ride and a few beers.
I think it might be a yearly family trip for us moving forward.
