We’ve seen this all before, of course; but this time seems different. There were platforms, then apps, then crypto, then Web3, with a sprinkling of EVs and biotech and medtech and agritech and energytech in between. 

I’m talking about the money meted out by venture capitalists over the last 10 or 15 years as they moved in lockstep to bet on some hot new sector. Mountains of cash spread around at the beginning of a next-big-thing. And then some winners emerge, and the rest recede and the VCs move on, looking to the horizon.

Which, of course, is AI. What is different are the numbers. They are big. I love huge numbers and exponential curves. Especially when they are new and shiny and have limited lifespan. And then I wait for the inevitable sag and enthusiasm wear-off and the shattering of overheated expectations. 

I do not mean to diss the VC community (many of whom I know). They know what they are doing, mostly. They only need one hit out of many to make their returns, so they can afford to make many promiscuous investments. One breakout hit pays for the rest and then some. 

But AI. I dunno. This feels different, so I hesitate to throw prognoses around. 

Consider that our search for AI stretches back to at least 8000 BC when – as we are told by pre-antiquity Greek mythology– there was a giant automaton made of bronze to protect Europa in Crete from pirates and invaders. He circled the island’s shores three times a day.

AI innovation has continued in a somewhat unbroken line since. And then in 2017, a couple of Google researchers released a paper titled Attention Is All You Need describing a new-fangled piece of statistical magic called a “transformer”; a radical new upgrade to “neural network” architectures, which had already been on the boil for decades. 

Which begat ChatGPT on 29 November 2022. (The ‘T” in ChatGPT stands for transformer). 

Which begat some very big numbers. 

But before we get there, let me dazzle you with this. The term AGI refers to Artificial General Intelligence, describing some as-yet-out-of-reach future in which a machine will be able to perform, reason and create as well as humans across a broad and undefined spectrum of tasks. 

As opposed to boring old traditional AI, which is good at chess. Or face recognition. Or auto-driving cars. Or cancer detection. Pfft. How one dimensional. How limiting. 

So, a research paper coming out of the Computer Science Department at Cornell titled Sparks of Artificial General Intelligence: Early experiments with GPT-4 was released on 13 March. It is 150 pages long and was written by no fewer than 14 researchers. In it they went digging into GPT-4, the latest release of the machine learning architecture on which ChatGPT rides. They analysed what it does well. What it does poorly. What it does differently from previous AIs. Where its borders are. What is a parlour trick and what isn’t.

And they conclude this:

“We demonstrate that, beyond its mastery of language, GPT-4 can solve novel and difficult tasks that span mathematics, coding, vision, medicine, law, psychology and more, without needing any special prompting.

“Moreover, in all of these tasks, GPT-4’s performance is strikingly close to human-level performance, and often vastly surpasses prior models such as ChatGPT.”

They continue: “Given the breadth and depth of GPT-4’s capabilities, we believe that it could reasonably be viewed as an early (yet still incomplete) version of an artificial general intelligence (AGI) system.”

Allow me to repeat that – it is an early version of artificial general intelligence, which wasn’t supposed to arrive for a couple of decades. 

Have we just witnessed the arrival of Gutenberg Press 2.0? I don’t know, but here is a consideration that surely has not escaped the notice of venture capitalists – this is the very first technology in history to learn. It can better itself; no other technologies ever invented can do that. And it will soon be able to do so both autonomously and exponentially. 

That has financiers reaching into their pockets to throw money at almost any start-up with an A and an I in its pitch deck.

And the big numbers and exponential charts I love so much? Here you are:

• 1 million people signed up for ChatGPT in its first week. A record for any software release.
• 100 million people signed up in the first two months. A 10,000% increase in a month. A big number.
• $500-billion will be deployed in AI by global businesses in 2023. I have no way to even place that number, but it is an order-of-magnitude increase in a few years.
• More than $40-billion will be thrown at generative AI projects in 2023 – 20 times what was spent in 2022.
• Japanese VC company Softbank is raising a $100-billion AI-only fund; another big number I can’t quite comprehend.
• GPT-4 is supposed to be 100 times more powerful than GPT-3, by at least one important metric (called parameters). And one wonders about GPT-5, because it took only a few months to leap from GPT-3 to GPT-4.
• ChatGPT scored in the 10thpercentile on a slew of advanced placement (AP) exams in the US, like AP Biology.  Now, three months later, GPT-4 scores in the 90th
• ChatGPT answers were restricted to 500 words when it launched. GPT-4 can now be requested to output 25,000 words. That’s novella length.

It goes on like this, startling stats spewing out daily. And this from a baby, born publicly a mere three months ago. 

A tsunami of innovation (scores of new announcements every week), exponentially improving technology, no end in sight.  

I fear I have no pithy conclusion to draw for you, because I am feeling overwhelmed by big numbers and steep charts. DM

Steven Boykey Sidley is Professor of Practice at JBS, University of Johannesburg.