That is exactly the reason LLMs are in a bubble that is practically doomed.

The horse to vehicle transition is actually a good analogy. When society moved away from horses, people bought cars. The companies that sold cars, if they weren’t completely incompetent, made instant profit off the sale of the vehicle. The oil companies made profit off the sale of oil as a lubricant and oil byproducts as the fuel. Companies and private individuals that used to use horses for transportation of goods and for production had real benefits they could see instantly. A horse requires large amounts of food that must be stored or purchased regulars. It is an animal that has physical limits to what it can do. A horse can only travel so far in a day before needing food, water, and rest. The amount of goods you can transport via horse team is limited by this. A car or truck needs fuel like a horse needs food, but refueling takes minutes not hours, and a single engine can replace entire teams of horses, making the transport of bulk goods far more efficient. The car can use the exiting roads for horses, less efficiently overall but still more so and faster than a horse. This efficiency and speed of transport had real benefits seen on paper practically overnight and led to massive infrastructure investment to take advantage of the vehicles advantages.

An LLM can initiate human writing patterns. That’s it. That’s all they are good for. It requires massive infrastructure advancement to even begin to work, and can only really replace communication and other largely writing tasks like programming. Although with any skilled task, even customer service, there still needs to be a human making decisions and ensuring the output matches the goals. So where a single engine can replaced 10 horses that you can then sell or dispose of because you bought an engine, an AI can replace a few people, but you still need so many experts to verify its doing the job. Imagine if every truck had to have a horse walk next to it to makes sure it was doing the horse’s job correctly.

By the time investment in roads and fuel infrastructure were made, vehicles had already been shown to outperform horses in every way. Yet LLM companies are putting the cart before the horse. They’ve invested heavily in the infrastructure for their products, and haven’t yet found any role in which they are more profitable or better than a human. So they’ve partially reduced the cost of their services and are operating at a loss. Can they sustain that until every programmer or customer service representative has died? Highly unlikely. If LLM companies fail tomorrow, 99% of the staff that was downsized can be rehired fairly rapidly. It will lead to a massive economic shift that will almost certainly trigger a depression, but the total number of humans that can still do the job the old way is practically unchanged. Whereas after over a hundred years of vehicle dominance, you’d be hard pressed to find enough horses on the entire planet to replace every vehicle in a country the size of America.