I don’t think LLMs are particularly useful for this. I can imagine that they might use them to write targeted social media posts, but there are better ways to analyze data. Maybe someone will use a LLM to comment/complete their code, but will this be reported in the media?!

@Daconomist in that case I think you may be more interested in the following market than this

I'm confused on what counts as a use here. A LOT of statisticians use co-pilot or gpt to facilitate writing/commenting the code for their predictive models. Does this count? Or must the LLM be actively suggesting the modelling approach? (presumably no, and yes respectively, or this seems way underpriced)

@NKM agreed, no and yes respectively

@firstuserhere i remember a paper from MIT (?) About how LLMs are really good at predicting how a certain message will alter a person's response to a survey question. Which is what inspired this question.

@firstuserhere Ahh, got it! I had something very different in mind; when I read voter prediction my brain went strait to the idea of using an llm to create models for forecasting results, not for trialling how well voters may respond to certain messages. Super interesting though, I'll see if I can find the paper :)

LLM's are awesome, but they are not the best tool for such perdictions. From the formulation of the question I also assume that this would resolve NO if the company was using an LLM, but ot directly in a perdiction model. So for an example, an LLM based ACE (Artificial Cognitive Entity) that is using an external tool for the perdictios does not count as YES.

@Swordfish42 yep, agreed on all points.

@Swordfish42 btw just to give motivation behind question- it was about how well could you model a human? We already have so many systems (ML based also) to do voter prediction, it was about whether we can take those systems, somehow do magic with LLMs throw into the mix and see if the ability to model and predict human behaviour increases