Framework for AI Skepticism

There is ridiculous, beyond religious hype about artifical intelligence (AI). There is also a ridiculous, beyond religious doom narrative about AI. Somewhere between those extreme views are tools that are useful for some applications.

I am developing an evolving framework for effective AI skepticism. This will help you avoid expensive mistakes by letting other people make them and leverage the benefits of these technologies when they are proven.

If I work with you to get your AI story in order, our journey will look something like this:

1. Private, local LLMs.
   An LLM running on your PC is as knowledgeable, but not as annoyingly loquacious, as ChatGPT, Claude, or any other cloud-base LLM service.
2. Chat is an illusion.
   Chat is also a hindrance. You need to work with LLMs using a completion user interface or API.
3. Embrace the bullshit.
   LLMs are good at knowledge, but are not good at facts. You need to understand the difference. What most call bullshit has a valuable knowledge component. You need to embrace that.
4. Beware of oracles.
   LLMs are not good at providing definitive answers. They are good at asking questions.
5. Beware of agents.
   LLMs are not good at doing end-to-end work. They can be good at identifying work for people to do.
6. You must lead.
   People in positions of authority in the market, your place of employment, your industry, and the government do not have the time or inclination to understand and embrace these subtle lessons. You will know that you understand the lessons because you will feel like you’re talking to walls when you talk with people about them. Monotone, untextured walls. It is your responsibility to lead by being effective where others are not effective.

These are not list items to memorize. They are things you will experience and internalize. You will not know them. You will feel them.

I am here to show you how these systems actually work and how you can best use them to your advantage. I am not here to validate your misconceptions — either hype or doom — about AI and LLMs. This is a very tough hurdle for a lot of people.