What LLMs Might Teach Us About Getting Unstuck

Large Language Models (LLMs) which underpin tools like ChatGPT, Claude and Gemini work by continually predicting the next token (typically a word) in a long stream. That is, the text (if that is the medium) of what they already started to write/say, will greatly influence what they write or say next.

It struck me today that probably our brains has some similarities to this process (that is definitely part of my experience), and thus perhaps we can learn something from this.

That is, sometimes when we are “stuck” and have a hard time working productively forward, we might need to set up a little bit of structure that makes it eaiser to “just think the next thought in the chain”.

As a concrete case, today I was very mentally stuck on how to proceed with a manuscript I’m writing. One of the problems is that the text I had already written was too large to keep in context in order to automatically see what I should do with it. While I did use an LLM to point out some problems with the scope and approach, I had a hard time finding an efficient way to work with the ideas myself.

But then, while getting more coffee (in itself a great way to get unstuck), I realized the quite obvious point that I should need to write down a short bullet-point of the main points we are trying to make with the paper.

Writing this in bullet point form makes it much easier to keep these things in context, and by keeping them in context we can much more easily reason about them, such as making sure we have covered everything important, and seeing how these points link together to create a story etc.

This was an example of cueing myself with a specific structure I could use for writing down the main ideas.

I can see how other types of cues might help at other times though, such as asking the right questions. Perhaps writing down a bullet point list of questions that need to be answered(?).

One can probably brainstorm a lot about different ways to cue, and different ways to structure one’s writing and thoughts.

Overall though, this reminded me about how cueing ourselves with the right structure or questions, regardless of how simple the cues might be, can probably help us getting unstuck a lot of times.