You'll regret betting against LLMs
GPT4 was released two weeks ago. I’ve been using it since it was released. It’s slow, it still hallucinates and frequently forgets the thread of the conversation that it’s supposed to remember. There’s been a lot of chat - including from me - that it will never succeed at ironing out those problems. It might not. But it would be a mistake to bet against it. OpenAI’s large language models are a classic example of disruptive innovation and disruptive innovation has an annoying habit of winning.
Clayton Christensen was the first academic to codify disruptive technology in his book Innovator’s Dilemma. It was a best-seller because it took the reader through the counterintuitive story of how companies could be well managed - expanding profit margin and revenue - but ultimately still lose. He showed how disruptive technology often starts by providing a simple solution to a problem at a lower price point than existing offerings - but with far inferior results - and then gradually improves its capabilities to the point that it becomes a viable alternative to incumbents in the market.
These disruptions are deceptive. They don’t look that disruptive. Or at the very least like something that will never get beyond a small segment of the market that’s driven by price, or a very specific use case. It means dominant players don’t see any reason to compete, until it’s too late. It’s why the lifespan for companies on S&P500 keeps falling and why massive companies like DEC, Kodak, Blockbuster, International Harvester and Sears have ceased to exist. The inability to adapt to disruptive technologies have seen entire market verticals disappear. And the latest disruption is in our digital space.
One story I’ve been repeating is from almost 200 years ago when railways were the disruptor of the day. During the First Industrial revolution canals were built across the UK. Freight could suddenly be moved almost all year round at relatively low costs. In 1825 The Stockton and Darlington railway was opened. Unfortunately for canals, trains were an even better innovation for getting things from A to B. Not only were they faster, they were less impacted by bad weather and they were substantially cheaper according to Michael Robbins in The Railway Age. If you’re delivering content, or experiences, you need to make sure you’re riding the train not the canal boat.
The other story I’ve been telling over the last week is the transition from large steel mills to “mini mills”. Using steel smelting to talk about the potential digital change is a little odd but it works. Mini mills like Nucor, Chaparral and Steel Dynamics now dominate steel production. But when they started they were tiny, tiny players. They used cheap materials and could only create the lowest quality, highest commodified steel called rebar. It’s the sort of steel that reinforces concrete and has terrible margins and very little value. Traditional steel companies were happy to be out of the rebar company because it was a grim, dog-eat-dog, commodity. And they couldn’t see the threat created by these new manufacturers creating bad steel. Over time though the output quality of these “mini mills” improved and slowly but surely they were able to move up the steel ‘hierarchy’ even finally figuring out sheet steel.
Transformer models like GPT4, or Google’s Bard, have the ability to move in the same direction. When GPT2 was released in 2018 it barely registered outside of the tech industry (and even there it didn’t make much of a splash). GPT3 made more of an impression in 2020 but there weren’t many people seriously considering that it could replace large pieces of labour. Fast forward two years to November 2022 and there were many more folks sounding the alarm bells. Each time it has made a step change improvement. It’s still deeply imperfect but it’s definitely got value and is getting better.
As small organisations, or nonprofit organisations, it’s sometimes easy to look the other way when there are case studies relating to historic innovation disruptions. Invariably they’ll be talking about massive, multi-billion dollar, corporates who were out to maximise their profits. It would be a mistake to take that approach. If you’re in the business of communication GPT-4 is here to take your lunch.
From our reading of innovation history these are the recommendations for what’s required to successfully move with the paradigm shift.
Commit with speed and agility
Companies that are able to adapt quickly to changing circumstances are often those that survive or thrive disruption. They are able to pivot their strategies,adjust or find opportunities to their new reality. And they do this by taking a decisive approach that is reviewed regularly as new information becomes available. In The Lean Startup Eric Ries advocates an approach that enables validated learning, where your organisation is able to take small, measured, risks to gain more confidence in the direction of travel. It’s about moving fast but making sure it’s possible to change direction of travel. This is not a time to adopt a ‘wait and see’ attitude, it should be about ensuring you work with organisational agility.
Companies that survive and thrive through disruptions are often those that are willing to embrace innovation and take an entrepreneurial approach to create more value for the people you serve. As Joseph Schumpeter, an Austrian economist and innovator said, “The entrepreneur always searches for change, responds to it, and [treats] it as an opportunity.” Now is the time to take risks and try to stay ahead of the curve.
Stay deeply customer focused
Successful companies prioritise their customers’ needs and preferences. They listen to feedback, empathise with them and unearth insights to improve their products services, business models or organisations. In The Corporate Startup - which is as applicable to charities and nonprofit sector, Tendayi Viki, Dan Toma and Esther Gons talk about how success is achieved when a company designs for the needs of their customers, not the needs of the company. This is still true at big moments of change. Understanding the shifting needs, problems and behaviours of customers is essential to move with this change.
Talent management and capability building
Companies that are able to attract and retain the best people are often those that thrive through disruptions. This is a pretty obvious thing to say: who doesn’t want the best people in their companies? But sometimes the obvious things are worth repeating. Richard Rumelt in Good Strategy/Bad Strategy talks endlessly about good strategy being easy to think up but hard to execute. This is an example. Investing in employees’ development and creating a culture that fosters innovation and creativity is hard, but it’ll pay dividends in the face of disruption.
A company who got this right was Intel. Twice. First with moving out from DRAM. Then - and even more impressively - when they brought out the Celeron chip, a cheap product that was ideal for the low-end PCs that were new at the time. Within a year it had captured thirty-five per cent of the market. They deliberately disrupted themselves with a cheaper product that solved a specific problem at a lower price. They embraced innovation, put their customers at the centre of their thinking and moved quickly with their teams to change their business model.
All of the above is to take a very defensive view of the situation. That’s not unreasonable. Kodak, Blockbuster, Pan-am, Borders and Tower Records would have all been smart to have taken a defensive view when their worlds were being disrupted. But It’s ignoring the positives that come from the “creative destruction” of paradigm shifts that Joseph Schumpeter bangs on about. The potential of Large Language Models is almost unfathomable. They could allow humans to unlock new ideas, communicate in different ways and experience art and audio we’ve never dreamed of. Yuval Noah Harari, author of Sapiens, sums it up nicely, ‘If we think about art as a kind of playing on the human emotional keyboard, then I think AI will very soon revolutionise art completely.’ We’ll be talking about this in a future article.