The Prototypes
kbbl - letting LLMs negotiate the government
After the Swedish general election in 2026 the parliamentary situation is complex. The race between the left and right bloc is very tight and the parties have constraints on whom they want to negotiate with. The last two elections have both lead to long, and complicated, negotiations before a new government has been formed.
Negotiations are de facto non-deterministic, same inputs can yield different results depending on who talks to who and when. This makes it very hard to predict, especially in cases similar to the Swedish parliamentary situation where the race is close and there are constraints down to the personal level of the people involved.
I recently listened to the podcast The Wargame and got fascinated by the mechanics of performing a scenari-based simluation so I thought, is it possible to do the same with negotiations for forming a government? This is what this prototype attemts.
The Swedish parties’ policy positions are codified and provided to agents who will then negotiate to try and form (or block) a government by trying to reach a platform that is acceptable to all sides. What’s interesting is that using LLMs for this gives exactly the kind of non-deterministic behavior that a real negotiation would have.
The challenge is of course if it’s possible to get the LLMs to act in similar ways as a human, so far the agents seem to keen on actually reaching an agreement…
There’s no demo to run but if you want to play around with the code, it’s available on Github.
Wayther - weather on the road
Is it only me or is it not a short-coming of regular weather apps that they only provide the weather for a period of time for a single place at a time. What if you’re constantly moving, then you’d like the forecast to be for the location where you will be at a given point in time, right?
I’ve had this problem in my head for a long time and the only solution has been to check the forecast for a few different places one by one to see what the weather will be where I will be at a given time. So sort of like the love-child of a route-planning-app and a weather-app and I bet you wanted one too (even if you didn’t realize it until now).
Enough said, here is the nice working prototype, an app that shows the weather for the route that you’ve planned. A forecast that is not static to a place but shows the weather along the route you intend to follow. Check out Wayther and feel free to fork the code.
The ToDont-list
You may have heard about todo-lists, things that you or a team plan to do organized in a list after priority. A todont-list is the opposite, it’s a list of items that you plan to either not do or stop doing.
It may sound silly (and well it is to some extent) but it’s a recurring theme in almost any type of work that you or your team are doing things that maybe you shouldn’t or that someone else should do. It can be cases of bad habits either personal or in a team or items that are taking too much priority because other parts of an organization is pushing them. In these cases it can be a good idea to collect those things on a list and track them to make sure that they don’t get done.
The todont-list follows the same logic as the todo-list but opposite. Items start as checked and your goal is to uncheck them. A todont-item can have something similar to a deadline but for the todont-item it’s called an avoid-until-date. An item that has an avoid-until-date set can’t be unchecked until that date.
If this is silly enough that you want to test it out, check out the prototype or fork the code on github.