Apologies for the lack of posts recently—life got in the way. I should be back to a normal publishing schedule now.
My view of the AI market in April 2024.
First, this is focused on LLMs. I'm ignoring tech like convolutional neural networks, etc., because I'm not as familiar with them as I am with LLMs.
Second, this ignores consumer applications, as I don't have any great insight into consumer behavior.
Foundational models--LLMs--are commoditized tech. OpenAI/MSFT, Goog, Meta are well-positioned here, given their large balance sheets. They can finance training and user acquisition costs in a way that independents cannot.
I'm not convinced that open source vs closed source matters as much as some people think it does. Both can thrive in this market. Akin to Windows/MacOS & Linux. Different use cases.
GPT wrappers have had a moment but they don't seem to have much of a moat.
I think of LLMs as a kind of substrate on top of which various apps, products, services can be built. These will primarily use AI agents to facilitate whatever it is that the product is meant to do.
The most fruitful areas will be vertical applications for a given industry niche. Horizontal plays are much more difficult to pull off, because then you are trying to be everyone's friend, rather than getting close to a few customers and becoming an industry leader for a particular industry. Life's much easier when you don't try to be everyone's friend.
AGI is probably a thing, and it will probably happen at some point in the future, but I'm rather skeptical about claims that it will happen within the next several-to-ten years.
Tradcorps--basically any corp not driven primarily by technology--will be slower to adopt AI than many technologists expect. But their adoption will be quicker than, say, cloud computing.