The Practical AI podcast interviewed the Fireflies CEO very recently (that’s who this article is about)
He was super transparent about being a cheap transcriptionist in the beginning, because he was trying to prove market demand in 2017, long before ChatGPT turned the world over. Basically doing it for his friends.
As someone in the tech and startup fields, I can say one of the best ways to make a great product is to do the thing you want to automate and learn how it works. This is an excellent example of that approach.
The co-founders held their tech startup together for 3-4 years, doing the best they could with keyword matching and voice transcription from the olden days, before it exploded in 2021(?) up to about 10m in revenue. Then they got early access to gpt 3.5 in 2022 because one of their investors was also an investor in OpenAI.
They were almost ahead of their time, and they were well positioned to take advantage of the huge power that GPT models brought to their business.
Excellent example of being in the right place at the right time, with a great vision and good network.
A decade ago my company was selling some “ML classification” service for service request calls, which in fact was 200 people in India being paid low wages even for India to do that 10 hours a day.
They were replaced by a simple neural network and I see they advertise now advanced AI agents, which I’m pretty sure are again 200 people in India
Nah it’s AI. Actually Indian. Amazon got popped for this last year.
“Amazon Mechanical Turk” works off of this
This starting to reassemble the dot com bubbie.
The AI stood for Actual Intelligence
A couple years ago I worked for a robotics startup that I learned got their start by 100% faking a demo for early customers. They had some magic box that an item would lower into and then the device would “scan” the item and spit out a bunch of data about it. It was entirely fake, the data was predetermined. That landed them a few early series investments that allowed them to do some “real” work. The founders were proud of this sham enabling them to start a company. Whole thing made me sick.
Reminds me a bit of Theranos. “Pay no attention to the man (or woman) behind the curtain”…
https://en.wikipedia.org/wiki/The_Dropout is a good mini series on it, I wasn’t overly familiar with the story
Except for, you know, admitting to fraud in public.
It isn’t fraud if all they promised clients was a note taking service. Contract wording is key in any scam.
“We told our customers there’s an ‘AI that’ll join a meeting.’"
What matters is what the contract says, not what they type on LinkedIn.
AI stood for An Intern.
It was just his name. Names Alan, Al for short
Article says it was the two co-founders taking notes themselves.
I think he was just trying to recreate Searle’s Chinese Room.
This concierge approach is nothing new; it existed even before LLMs are a thing.
One of my courses in undergrad computer science is Human-Computer Interaction, in which we learn about user experience (UX) concepts.
One of the things we learnt is to validate our ideas quickly and cheaply before putting a lot of time, effort and money into building the thing.
To do so, what we can do is build prototypes. The early versions may be be low-fidelity (lofi) and are scrappy. The later, high-fidelity (hifi) ones would mimic the functionality of the actual products, and may even appear to work to end users when in reality it could be just be manual effort behind the scenes.
The example given during lecture is the development of a ticketing system. To test the idea out, one could simply get a dude to sit in the “machine” and give out slips of paper.
Anyway, I am explaining all these because this seems like a surprise to those without the same educational background. Long story short, what this startup did is completely normal in the realm of software.
We may have better tools like Figma to simulate browser / mobile frontend experiences, but nothing is stopping us from going back to the basics and doing it this way.
I get it, you enjoyed your HCI course. I did too.
What you need to understand is you can do this, but you can’t mislead investors into thinking your Mechanical Turk isn’t a man in a box. That is fraud.
I can’t stress enough: this is literally why Elizabeth Holmes is in jail right now.
Figma balls lmao
I don’t think that’s anywhere close to what was happening here, though.
There already are transcription services (which are now being branded as AI but are using tech that has existed for 15 years at least). So, surely this startup’s claim is that they’re going to make a more advanced transcription service using new AI algorithms. And as a “proof of concept” they secretly use a human to perform better transcriptions than any existing algorithm does. So in reality it proves jack shit.
Their interface is just a rehash of existing services, the only thing new they could bring to the table is an actually better algorithm, which they don’t have, so they just faked it. Meaning they have nothing of value at all.
Not just normal, but correct. Try out a workflow before enshrining it. Nothing is perfect from the beginning.
Ahh, no. If this guy’s company was in the prototype stage and they were doing this with potential, knowing clients then I’d agree with y’all. But this is someone selling a lie.
It’d be like if I said I was selling a multiplayer game, and when it came out, all the other “players” were NPCs with your friend’s username. Don’t worry, I’ll figure out the actual multiplayer later! This is normal and correct!
…all the other “players” were NPCs with your friend’s username. Don’t worry, I’ll figure out the actual multiplayer later! This is normal and correct!
Have you heard about the term
.io games? Because this is indeed something that happens regularly…But he was selling transcripts and delivered transcripts. The way they are generated is irrelevant to the client.
Dude’s winning at life. Providing the best possible service.
Hahaha this is brilliant.
Pay no attention to the man behind the curtain
We’re taking the jobs back AI!..
honestly…Yeah man I’d pay for this.
Fake it 'til you make it!
That way maybe some of it will be accurate.
I do find it incredibly expensive. Why prefer a ai offering if it ain’t cheap?
And it didn’t happen by accident either. That was his business idea right from the start. Even the eventual reveal.
(I don’t know, but I hope so)
