LLM Expectations
Today I tried to create with an LLM something I had a concrete idea in mind for.
I could not, within the timeframe and patience I possessed, get the LLM to do what I wanted it to do.
This is in stark contrast to my usual experience when working with LLMs, where it’s trivially easy to get them to do what I want them to do.
And it’s trivially easy because typically I’m asking it to help with things I’m a subject matter expert on. When it comes to 3d animation, and in a browser especially, I know basically nothing.
Even if I wanted to be more specific, I could not.
It’s frustrating to use a software tool I’m not familiar with, and I believe the same is true of LLMs. When trying to get things done in an area of knowledge the prompter is not at all familiar with, it’s not much different than something like not knowing how to use photoshop to do something specific to a photo.
Except, in the end, you can still get something kinda neat, despite being not at all like what you wanted.
I could have spent hours using the LLM to learn about different types of visualizations and 3d effects, and then prompted it to utilize them in a nice way. But that isn’t what I wanted to do.
And it’s bad at doing things the user is not directly guiding it to. It’s just a tool.