We are living through a time of extraordinary technological innovation that shows no signs of slowing; indeed many technologists believe the rate of innovation will continue to increase thanks to the powers of machine intelligence and other factors. Large language models (LLMs) like ChatGPT are a case in point.
“We can't escape ChatGPT ... I've spoken to more than a couple of dozen CEOs or other C-level execs in publicly listed companies. And that is the one common story that comes up each time about priorities for the year,” Azeem Azhar tells Pictet’s Found In Conversation podcast, hosted by Edie Lush.
“I think as an adoption curve, it will be faster than the adoption of electricity, or even the adoption of typewriters, which took about 20 to 25 years to have a real impact. But it will still take a measurable amount of time.”
Crucially, although the technology is clearly complicated, it is an area where small start-ups can thrive. Paris-based Mistral AI, for example, has developed a rival to Miscrosoft-backed ChatGPT on a shoestring budget.“
“I would say it is quite hard to build them de novo. But people are proving, as this French team and Mistral have shown, that small teams with far fewer resources can get there. And that, I think, is a common pattern in the history of technologies. It's extremely hard to get it to do it the first time, but as soon as someone's done it the first time, a few things happen,” says Azhar, who shares his tech insights with hundreds of thousands of readers through his weekly Exponential View newsletter.
Firstly, the developers know that the technology is possible, that the software can be successfully created – because it already has been.
“We now know we can do this, which is a great motivator for teams. The second thing is that, of course, some of the know-how leaks out and some of the strategies that they've used leak out. So the other teams can more quickly find ways that work. We are able to do get much more out by putting much less in with these subsequent models.”
He gives the example of GPT-4 – a multimodal AI model that can use images as well as text prompts to create content. This pioneering model was created using a training set of some 13 trillion tokens, requiring huge amounts of computational power. Subsequent models can achieve similar results much more efficiently.
“That is effectively like saying the motor engines of 1923 were more efficient and more powerful than the ones of 1896. It's just happened much more quickly,” Azhar says.
“The frontier will remain the domain of the very best teams, helped by the very best AI. It's hard, but it's going to become more and more commonplace.”
If you would like to hear from more experts on understanding the modern world, listen to the Found in Conversation Podcast.
Investment insights
by Stanislas Chanavat, Principal Thematics - Private Equity, Pictet Alternative Advisors
-
The emergence of ChatGPT and other forms of generative artificial intelligence has certainly captured the public attention and revitalised investor enthusiasm for technology stocks. We are clearly at the beginning of another major technology shift, one that will transform most technology (and non-technology) markets over time.
-
AI has the potential to be a major catalyst, unlocking new efficiencies, and creating attractive investment opportunities across companies that are pioneers, providers, and adopters. However, the long-term victors in such paradigm shifts rarely become apparent in the very short term, so some diversification of exposure is important.
-
For now, we focus on solid technology and defensible business models, avoiding the highly-priced and capital intensive companies. As generative AI matures, such companies will be well-placed to harness and monetise its potential – as they will be with the next headline-grabbing innovation.
