Holy potatoes, there goes the future.
If you haven't seen the demo yet - you've got to check it out:
The new flagship model: Spring Update | OpenAI GPT-o omni is here.
This can't really be real, can it?
I'm truly blown away by what Omni can do.
Open AI have taken another leap - a lightspeed leap - forward into the future, or what is clearly now, the present.
I actually thought that maybe some of the demo in the update was rigged - it's too impressive to be true. My first thought was, this can't really be real, can it?
Always a good sign when looking at new tech, unless it actually is too good to be true.
But I can't see that being the case. This is genuine.
Through this AI we're realising the dreams of the many sci-fi books, shows, and films of the past.
We are thrusting ourselves into what feels like a bizarrely premature future.
One of AI-powered anything and everything.
It's amazing, it's exciting, and I can't help but feel more than just a bit scared of how quickly this technology is evolving before our eyes.
As a technologist and enthusiast my entire life, no technology in our history has ever moved this fast before. And it seems to be accelerating.
(And I've seen a lot, starting from my 7-year-old curious self on my Dad's green and black Amstrad word processor, I've seen the start and evolution of Google, broadband and mobile phones).
The speed is scary. It raises a lot of questions about regulation and safety on a population level, but bloody hell is it exciting at the same time, and the potential is limitless.
First things first though, we need to address the elephant in the room!
Is it HER?
I can't be the only one thinking this - but is that Scarlett Johansson whispering sweet nothings to me as an AI bot!?
(For those who haven't seen HER - the amazing film with Joaquin Phoenix and Scarlett Johansson as his AI operating system - do watch it, it's brilliant).
For Omni, it's like they have used her soothing, mildly flirty (and let's face it, sexy) voice, and made it their own.
(I don't actually think it's Scarlett's voice - but with AI these days - who knows!)
My Hollywood crushes aside - this demo is a snippet of things to come, get ready for a new wave of voice bots that will make our Alexa's look like total numpty calculators compared to an iPhone.
Human Emotion and self-awareness
The last part in the video (and this is a spoiler - I really recommend watching it) - is where she (and yes I've started to refer to chatGPT 's Omni as a female) asks Barret (one of the demonstrators on the stage) to see if chatGPT can tell the emotion from his face.
(See 23:46 in the video)
He smiles, gawkly.
She captured his emotion, which was cool. (To be fair, it was a pretty exaggerated smiley face he was pulling), she also captured the added excitement, which was awesome, and bizarrely accurate in the context of the stage.
But then, the mic drop moment for me, when they said they were discussing her capabilities.
It actually made me feel quite emotional when the response came back.
Probably a culmination of what I had seen so far, the capabilities, but also, she responded to the praise with a bit of humility, or modesty maybe - that was really good.
This semi-self-awareness bit- or the response suggesting it, was the most human-like part about the display. And yes, whilst my Alexa can give me programmed responses to me giving her compliments, this felt more real. It felt like a connection.
The connection doesn't have to be real for it to be felt, we've all been there I'm sure.
But it's that connection that has been missing from all these chatbots so far.
They have been pretty dumb automatrons up until this point..
This illusion of self-awareness, and the illusion of connection, it's the game changer we've been waiting for I think. It's going to be what sells, mark my words.
Now I know it's not actually self-aware.
I understand (to a degree) the technology that LLM's are built on.
But this was not a programmed response. This was a response generated by selecting an appropriate context, and building a chain of words using the transformer model and forward and backward propagation and so on so forth. It was a machine learning response, but it was a good one. It felt real, and it was as good or better than most humans can manage when given a compliment.
Coupled with the ability to get things right of course, which omni can seemingly do better than most humans, something clicked with me watching this.
I can honestly say I haven't been this excited about a piece of technology for about 20 years. (Exactly 20 years actually, the last time I was this impressed was when the Nvidia Geforce 6 series came out and I benchmarked it on Doom 3 - it smashed it).
So much so, I'm thinking hard on how I can use ChatGPT for my medical endevours.
The Future
I'm excited about where this technology is right now, it's jumping in leaps and bounds, but more importantly, where it is going to be in the very near future...
Not just what we're seeing on the surface, but what's being done underneath the hood is truly an evolution of our technological capabilities.
On what they showcased in the demo:
The speed of processing, especially with the new realtime translation abilities is incredible, and the practical use of that is immediately obvious, especially in clinics.
The ability to read and reason paper-written equations - is mind-blowing.
The interpretation of the graph. Holy potatoes.
Think of the potential for every industry out there. Think of the potential for every student, every person - and every patient.
It's literally bringing the world's knowledge to everyone, the difference this time, it's on a shiny platform / platter, and it has a nice voice to round it off.
Summary of now
Maybe we need to slow it down a bit and maybe we need to start doing more to regulate this technology.
Maybe general AI is a lot closer than we previously thought.
Maybe this is just a really good party trick and underneath the magic is something not as impressive as it seems.
For today though, I'm impressed, excited, and looking forward to playing with the cheaper Open AI 4 Omni API that I'm going to plug into my Llangchain right now.
