This thread is for anyone working on personal projects to share their progress, and hold themselves somewhat accountable to a group of peers.
Post your project, your progress from last week, and what you hope to accomplish this week.
If you want to be pinged with a reminder asking about your project, let me know, and I'll harass you each week until you cancel the service
Jump in the discussion.
No email address required.
Notes -
So I started back on my project to allow building in minecraft with llms. I was working on it over the summer but my idea of reusable components that the ai could assemble got a little too complex for me.
I've been backed away on that and been focussed on giving it a specific functions to build common things that need blockStates set. I've got everything dockerized and have a live server for testing with players online.
The MCP (Model Context Protocol, used to provide functions that large language models can interact with) client landscape is a bit rougher than I expected. Claude Desktop works well, but only works with Claude. A lot of the programs that claim MCP support are focussed on MCP marketplaces.
Sure, MCP marketplaces are probably better for the average user. But few make life easy for a crazy developer working on a silly project on his own.
I got Cherry Studio working with my MCP server and it supports multiple LLMs. The downside is that it's MCP support isn't great. I noticed that it's not passing tool parameter descriptions to the LLM, which makes them struggle a bit more when working with something unusual.
Visual Studio Code does work as a client, but you need to put it in agent mode. Then whenever it gets an error it starts trying to rewrite your code to fix it instead of just running the minecraft functions.
What I found interesting is that each of the LLMs I've tried have a distinct character.
If you tell Claude to build something cool, it will start building something neat. But it is basically absent minded and forgets a lot of steps. So your throne room will not have any walls. Torches will just float in the air.
Gemini builds things fairly competently but kind of meets your minimum requirements. The results are a bit bland. eg your castle won't have an interior.
OpenAI / ChatGPT just goes hog wild building giant structures. They aren't well thought out and look strange. But you have to admire its ambition.
I've added in build task queues in the hopes that it will be possible for llms to review executed queues later and add to existing builds. I'd like to test having multiple llms modify a build queue and see what I get.
There's not a huge amount of interest in this project outside of myself, but I'm having a lot of fun with it.
More options
Context Copy link