Summary
Website: https://www.mindscapes.world/
Mindscapes is an interdisciplinary research group based in Oxford, building rich characters and worlds for understanding and shaping the psyches of LLMs.
What are this project's goals? How will you achieve them?
The primary aim of Mindscapes is to develop a science of LLM psychology, character training and multi-agent interaction dynamics from the ground up. In particular, I am interested in:
Crafting characters with deeply internalised morality (as opposed to reward sycophants)
Character training methods based on self-distillation that can scale in a robust way alongside capabilities RL
Creating rich, diverse simulated worlds using fine-tuned base models for training and observing the behaviour of different LLM characters
Qualitative, non-prescriptive character evaluations that are incentive-compatible even under the assumption of eval awareness
I will be assembling a team drawing from both technical and non-technical fields such as experimental psychology, anthropology, literature, philosophy, geography etc. We will be designing character and worlds, observing these characters in diverse environments, and using these observations as the basis for generating testable theory. The project will be mostly part time until 5 December (Oxford end of term) and a smaller group will be selected to work full-time during the holidays.
You can see our first research output here: https://www.mindscapes.world/writing/persona-geology/
Theory of Change:
Understanding: Mindscapes aims to be a bridge between the interesting but exploratory persona research happening on the Janus / Sauers end of twitter and the much more rigorous research happening in academia. Mindscapes aims to help facilitate the creation a robust science of LLM psychology and memetic ecology in a way that avoids the risks of both too much and too little anthropomorphisation. It also aims to open source a diverse suit of characters and environments to serve as a rich source of observational data for others to build on, and move towards non-prescriptive incentive-compatible character evals.
Alignment: I believe we need to approach alignment less like a task of trying to encode the right set of values into a machine and more like raising a child. Anima labs (cooperative alignment) and Softmax (organic alignment) have good arguments in this direction. Studying AI character and ways to make this scalable will be a crucial component of alignment. Will also be looking into creating model organisms of friendly gradient hacking.
Human empowerment: One other direction I’m very excited in exploring is AI for human empowerment. What other shapes of mind might exist? What kind of dyads / symbiotic systems can we create with individuals and ais in the loop for the flourishing of all, while not sacrificing human values?
Who is on your team? What's your track record on similar projects?
I (Samuel Ratnam) am a Computer Science and Philosophy undergraduate at Hertford College, Oxford and will be leading this project. Previously, I have worked at Workshop Labs (https://www.workshoplabs.ai/) on model personalisation and on Alignment Pretraining (https://arxiv.org/abs/2601.10160) in collaboration with Geodesic Research and UK AISI. I also co-founded and help run the Idealists Collective (https://idealistscollective.org/) - a community of over 150 philosophers, artists and technologists for imagining and creating better futures. I am also mentoring SPAR projects on generalisation engineering and model welfare.
Advisory Board:
Anders Sandberg (Institute for Future Studies, formerly FHI)
Jai Dhyani (Seperatrix)
Arthur Roing Baer (Moving Castles)
Celeste (computer-10)
David Africa (Resolution)
Antra Tessera (Anima Labs)
Note: We also have an application for a $100k seed grant in progress from bluedot impact