Think of AI tools as "apprentices"

Here's another interesting "experiment" with ChatGPT:

https://chat.openai.com/share/dfb5748f-151d-488f-a7f4-5134043c1189

?

Couple of interesting observations:

- despite having only 2 very simple requirements, ChatGPT was not able to identify an inconsistency/conflict between those requirements

- finally, with some "coaching/guiding", ChatGPT managed to "see" the inconsistency/conflict

- still, after providing it correcting information, ChatGPT "refined" the requirements with the same inconsistency in it (despite "saying" that it fixed it)

?

Don't get me wrong, AI tools like ChatGPT are excellent tools. But, naively thinking that AI tools act as intelligent "analysts" is still a "work-in-progress" reality.

A much better approach collaborating with AI might be to think that AI tools act as your apprentice (or, better yet, many of them).

Say, you are a very smart person. You also have a library with tens of thousands of books. You are trying to solve a problem that requires lots of knowledge, skills, expertise, intuition, and common sense and you know that there is lots of useful information in those books. But, darn it, you don't have the time or patience to dig for that information. Here are your apprentices coming to the rescue. You ask each one of them to “browse” through the thousands of books in your library and bring back a summary of the books that seem to be related to the problem you want to solve (of course, you guide them what to look for). The apprentices themselves don’t know how to solve the problem (or even how relevant the information they find is towards solving the problem). But you do.

Based on what they brought back, you ask them to dive deeper into some of the books (the promising ones) and to put the other ones (the “not so useful this time” ones) back on the shelves. You also ask them to read certain chapters (or the whole book) in the promising books and bring back the “next level of details" when they are done. When they do, repeat the whole process.

Eventually, you get the information you need to solve the problem and, with that information, you start coming up with a solution. And it did not take you 100 days to get all that information. With the help of your very helpful assistants, it took you only 1 or 2 days to get it. Also, you were smart enough to NOT ask the assistants themselves to solve the problem, since they can't do that (at least, not right now, not until they master your expertise).

Of course, this not the only way to use AI tools. But it might help you get some ideas on more effective and efficient ways to “collaborate” with YOUR AI assistants.