Insanity, LLMs, determination and the future of software development

Insanity Is Doing the Same Thing Over and Over Again and Expecting Different Results

The above quote may or may not be by Albert Einstein. It doesn't really matter.

If you've been around software development for more than about seven minutes, you've probably heard it quoted in some form or another.

It's often used as an argument against blindly repeating yourself, especially when trying to recreate a bug with vague repro steps.

This quote also points to a fundamental difference developers need to consider as the use of LLMs and "AI" become more a part of the software being developed (not just as a tool for developing the code.)

AI (& LLMs in particular) has (have) a level of non-determinism about it (them).

Repeatedly asking the same thing of an LLM shouldn't always produce the same result. 

There's a random element that you almost certainly can't control as an input.

Many software developers likely the certainty of software as (in theory) the same inputs always produce the same outputs. The level of certainty and determinism is reassuring.

Businesses like this too. Although it may not always be as clear.

Looking at the opposite situation highlights the impact on businesses.

Business: "How (or why) did this [bad thing] happen?"

Developer: "We can't say for sure."

Business: "How do we stop it from happening again?"

Developer: "We can add handling for this specific case."

Business: "But how do we stop similar, but not identical, things happening?"

Developer: "We can try and update the training for the AI, or add some heuristics that run on the results before they're used (or shown to the user), but we can't guarantee that we'll never get something unexpected."

Fun times ahead...