We’re currently in the “silly season” of hype over “Generative AI,” machine learning programs whose output can be text, code, pictures, audio, or video.
There’s a lot of experimentation and there are a ton of bad takes.
Here’s how it’s going to go.
There are two things to know.
First, any software requires training before it becomes useful. Right now programs like ChatGPT are creating garbage. Interesting garbage, but garbage. As people learn how to get value from these tools, however, there will be huge jumps in productivity. The timing and size will differ from industry to industry, and from organization to organization. But it will happen. People will be able to do more, high-quality work in less time. This won’t reduce the need for people. It will give people more things to do.
This is how it went during the last step change in software, SaaS databases. There was resistance, there was experimentation (often under duress). Valuable solutions emerged over time as people came to understand the tools.
Second, garbage in, garbage out. If your input is everything on the Internet, or everything in Google, it will be subject to limitations of falsehood and bullshit. If you want to get useful output, inputs must be vetted. This will create value for large volumes of truthful material. That value will also create a business model for vetting.
None of this is going to happen right away. The valuations being placed on many companies in the space, like Nvidia, are ridiculous right now. I don’t know when disillusionment arrives, so I won’t advise a short. But before revenue becomes real there will be a reckoning. There always is.
One final point.
Machine learning can increase productivity, but it can also create new capabilities beyond words and pictures. This is a big deal for the Machine Internet. If inputs to a system are vetted and secured, you can manage an entire city’s traffic system reliably. You can manage a hospital’s flow of patients and worker time accurately.
You can create systems, based on sensors and computers built on a single chip used on wireless networks, that would amaze people from way back in 2019.
This is going to be fun.
