Two questions for economists about LLMs.
In general I would expect technological change to produce deflationary pressure rather than inflationary pressure. The point of the new technology is to produce more utility for the same price. You might expect inputs to the new technology to rise in price, but this should be more than offset by falls in prices of substitute products. If the overall price level doesn’t fall, what’s the point of the technology?
However, LLMs are producing inflationary pressure. This is because there is so much expected future value from LLMs, they are justifying enormous investments in very specific technologies. GPUs are the obvious example, although not that interesting since the specific designs being used are not useful for anything except LLMs. RAM is the better example: a genuine commodity product essential for large swathes of the economy, and now shooting up in price due to monopolisation for investments that are not expected to pay off for many years. The implication is that perhaps decades from now the general price level will indeed come down, albeit likely leaving semiconductor products still high.
My question is simply: is this normal? My own experience of technology is continually falling prices, not just in the technology sector itself but also in other areas such as food. Inflation is confined to physically constrained sectors such as real estate or minerals. Technology normally moves at a slow enough pace that short-term demand changes for inputs can be absorbed in time to keep prices low. So perhaps this is an artifact of the sudden leap forward in LLMs.
On the other hand, this could be one more sign that we are taking on an unjustifiable systematic economic risk. Or to put it another way, the whole thing is a long con. There is another, far more significant technological breakthrough happening as the background to all of this: solar panels. And that technology has never been anything except deflationary. The vast number of solar panels being churned out do not seem to have measurably increased the cost of silicon wafers. In that context, the increases in RAM prices smell off.
So the other question is: what happens in the likely event that the expected benefits of the technology do not fully materialise as expected?
One story would be, many businesses are currently postponing investments, hoping that LLMs will mature and make those investments far cheaper. When it becomes clear that they will not, those investment plans will be unlocked, leading to a boom in conventional technologies.
The other story is that so much of our material wealth will have been locked up in useless data centers that the economy will crash. We will suffer a “hangover” from over-investment.
Of these two, the first makes more sense to me. I’ve been cautioned against economics-as-morality-play: making large systematic economic mistakes doesn’t require “punishment” from the money gods. Rather, disasters usually result in the need for everyone to roll up their sleeves and get to work. After the second world war there was a widespread economic boom. That effort was largely invested in replacing what had been destroyed, so it wasn’t exactly about material wealth. But at least unemployment was not a large concern.
But that seems an overly-optimistic interpretation for our stagnating 21st century civilisation.