[Me & X] Will AI take a long time?
X: Do you think it will take a long time to develop the practices around AI?
Me: A good few years. They are emerging however, which is why you should be learning, getting involved and experimenting.
X: So why do people think AI is going to spread slowly?
Me: No serious technologist thinks that. Well, dodgy consultants will try and persuade you that "others" think that and only they have the wisdom to realise that we're in a punctuated equilibrium but given most people went through cloud - everyone realises this.
X: What's a punctuated equilibrium?
Me: A property of industrialising technology, something which is happening to LLMs. It's the exponential nature that really fools us. I used this to estimate where AWS was going back in 2009-2010. Try reading chapter 10 - it has lots of useful bits in it.
X: No correction then?
Me: Why do you think that? It's very likely there will be a correction, your only question is whether it's loud (an analogy would be web 1.0/web 2.0 circa 2001) or quiet (think cloud, circa 2015 to 2019). The difference being whether it hits the providers and is seen in the stock market or whether it hits the customers and is more of an internal reality check (i.e. a painful realisation that someone else's computer isn't infinite and still obeys laws of economics and physics). We then move on to being more productive.
X: Will it be quiet or loud with AI?
Me: Let us be more precise - AI is a field and I think you mean LLM/GPT. These components are infrastructural for the future, so we have to distinguish between providers, start-ups built on the providers and general customers. Being infrastructural and pay as you go tends to hide corrections i.e. people don't go bankrupt overnight and instead bleed money slowly. On top of this the providers have real customers, real revenue and real lock-in. That would tend to favour a quiet correction for the providers with a plateau of market value (with some volatility) whilst revenues increase at a slower growth rate with a focus on margins. For general customers, you'd see rising bills and an internal focus on efficiency i.e. more questions and challenge over the spending. For the start-ups, a lot of slow deaths. What could make this loud is whether the financial market will lose confidence.
X: The market has overpriced these firms?
Me: The future value doesn't look unreasonable, it is a punctuated equilibrium and these are future infrastructural components. But, confidence is affected by many things including concerns over financial engineering and how real the customers and revenue are.
X: What will happen?
Me: Highly sensitive to individual actors actions.
X: What do you think?
Me: Think LLM/GPT as future infrastructure. It's big, real but not yet stable enough to justify maximal irreversible spend. I'm with Burry on a loud blip. Doesn't change long term trajectory hence learn, get involved and experiment.
Originally published on LinkedIn.