Interface and interaction
The prototype is a dense, unfamiliar body of information. Work here could touch information hierarchy, navigation, responsiveness, accessibility, visual design, usability, or Godot implementation.
Paid collaboration // small scale
Work with Mesocosm
A playable Godot prototype needs human craft, specialist judgment, and help becoming something people can investigate together.
Where the work begins
The slower kind
of knowledge.
Much of the current build grew through AI-assisted development, mixed with ordinary human effort and pieces carried forward from earlier experiments. It works. Mesocosm now needs people who have spent time learning how interfaces communicate, how software holds together, how visual systems acquire character, and how interpretability work earns trust.
Performance and systems
Profiling, loading, rendering, simulation scale, architectural cleanup, and the practical work of helping a stitched-together prototype grow without losing its useful parts.
Art and visual development
Art direction, effects, animation, asset work, environmental character, and a practiced eye over the whole. Mesocosm already has a visual language and plenty of room for another person to work inside it.
Interpretability and infrastructure
Conceptual advice, scientific review, reproduction and provenance, data retention, shared infrastructure, and the beginnings of a public commons around citizen-led exploration of GPT-2 Small.
Latest prototype // Godot
Inside the current build.
Layers become rings. Neurons become volumes. Replayed sentences leave routes through the model. The current build can be navigated, filtered, inspected, and extended.
Collaborators will be invited into the project to examine systems, develop ideas, and work directly with the code.
L5:N541 // seven volumes // thread view
Help make the inquiry stronger
Methods people
can carry forward.
GPT-2 Small is a manageable shared object of study. The value here lies in the practice around it: keeping observations, reproducing procedures, comparing results, and giving curious people a credible way into the work.
Mesocosm would like to grow into a small public data commons for citizen-led interpretability. The first useful version needs thoughtful choices about provenance, storage, contribution, replication, and the claims attached to each record.
01What should be retained so another person can reproduce an observation?
02How should runs, interventions, failures, and interpretations relate?
03What can the instruments teach honestly, and where do their limits begin?
04How can many small investigations accumulate into useful shared material?
Forms of collaboration
Start with a
useful piece.
Engagements can take the shape that suits the work: a few hours of specialist consultation, a defined commission, a short focused sprint, or occasional part-time involvement. Scope and payment are agreed before work begins.
01 // Access
Work in the projectWalk through the current build, examine the systems involved, and work directly with the Godot code when the contribution calls for it.
02 // Place
Remote by defaultMesocosm is based in Australia. Collaborators can be anywhere and most work can happen asynchronously.
03 // People
Many routes inProfessionals, researchers, students, artists, developers, capable hobbyists, and people whose practice crosses several of these categories are welcome.
04 // Scale
Small and paidResources are modest. A well-shaped contribution can begin small and grow when the work, timing, and available budget line up.
A useful first message can be brief: who you are, which part of the project caught your attention, links to relevant work, and what kind of contribution you can imagine making.
See the current paid prioritiesScientific review comes first. A multi-text data pipeline is the clearest engineering example downstream.→Open channel
Make contact.
If something here sits close to your practice, Mesocosm would like to hear what you noticed and where you think you could help.