In general, I find it very useful to imagine how things (very broadly speaking) may have gone were an event negated or another event introduced. Doing this systematically, even using standard chat models, proves to be quite difficult and generally unwieldy. To address this, I have created the Hauntoloscope. Under the hood is a prompt that allows Kimi-K2 (my favorite model for this type of thing) to generate news articles in timelines where a single event is negated or a fictional event has occurred.

I specifically think this will be of interest to the broad rationalist community. To that point, I am interested in examples where the prompt could be improved. You are able to import and export entire generation objects (called Relics) as .json files. These are easy to share. 

I would like the counterfactuals to be as realistic as possible (anything less would make it not very useful). Currently, I think the generations are ~pretty good, but there is certainly room for improvement.

Currently, it is bring your own API key. I am using Groq because it is incredibly fast. I have not monetized this whatsoever and am more interested in making it a useful tool for thought. It is also fully open source.

I am excited to see what people end up using this for!