I’ve been on the optimistic side of the aisle when it comes to AI. It’s hard to look at the potential business uses of AI and not see an explosion of productivity and firm profits. A few of the benefits I see are
I don’t think any of the above are out of the question at this point. With that out of the way, I’ve downgraded my expectations for AI as I’ve seen a number of things play out.
1. Frontier LLMs are very expensive. This has been documented elsewhere by people with more knowledge than me. The business case for LLMs rests on doing things that are cheaper than the alternatives. Not only are token prices high and increasing, tasks like vibe coding eat an extreme amount of tokens. Even worse, once you have vibed some code, you have a blob of code that requires a frontier LLM to modify and maintain, and a large code base eats a ton of tokens for even the smallest change. We have less information about LLM usage in other areas, but I doubt it’ll be any better than programming.
2. We have significant infrastructure limitations with respect to energy and computer hardware. Both of these are problems that have to be solved. They may not be solved.
3. Most business uses will require fine-tuning. Generic models are too limited in what they can do to be productivity-enhancing. That’s a different story from dumping text into a chat box and reading the answer.
4. Inference in the cloud is looking increasingly unlikely as an option for businesses. It’s going to get harder to convince businesses to send their data to the cloud, particularly given that (i) they have no way to know what’s happening with their confidential data, and (ii) even if they had evidence that something was going on with their data, they wouldn’t be able to do anything about it. The future is fine-tuned local models, and the “local” adds another layer of complexity.
5. The word “AI” has been tainted by actors with bad intentions. Guys like Dario Amodei, Sam Altman, and Elon Musk have been smearing dog shit on top of all things AI for a while now. Their, let’s generously call it “questionable claims” around AI, in the name of convincing investors that they can be part of history for a few million dollars, has made AI look like something a lot of people don’t want. Couple this with hatred of data centers (now the fastest way to end a political career), high electricity prices, high computer hardware prices, and the average worker being sick of talking/worrying about the topic, and it’s easy to see AI going the route of annuities. In spite of their potential benefits, many people avoid annuities in all circumstances due to their reputation. Once Gen Z starts taking over management roles, we may see strong opposition to AI because it’s AI. Younger workers will say “We should use AI because…” and the manager will interrupt them: “As I told you before, we are not using AI garbage. It only makes things worse.”