You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
I built a running field substrate with no objective, no loss, no reward, and no global controller — only local physics and accumulating history — and showed that identity emerges measurably from developmental trajectory. Accepted at ALIFE 2026 (LBA lb113). Currently under evaluation at the EA Long-Term Future Fund.
Goal: turn a peer-review-accepted, directionally-consistent result into a statistically powered, independently re-implementable research program on measurable individuality.
1. Months 1–2 — statistical power: scale n=8 → n≥40, publish full analysis scripts + ablation archive. 2. Months 2–4 — stakes: implement genuine metabolic dependency and irreversible collapse; test whether “mattering” can emerge from physics rather than reward. 3. Months 4–6 — transformer bridge: test whether memory-as-terrain (parameters rewritten by the interactions they govern) can be applied to frozen-weight LLMs via an external trajectory-recording space. Deliverable: a documented positive or negative result.
Budget: stipend $1,400/mo (deliberately low cost-of-living) · compute/API $300/mo · onetime GPU $1,500 · conference $700 · 10% contingency. At $5k minimum: months 1–2 only (statistical power + full replication package).
Full budget spreadsheet:https://docs.google.com/spreadsheets/d/1sbooBeCI02MPMteMpiNQFkV7tmZzNe4wUl-YwvHORXs/edit?usp=sharing
Solo, on a single second-hand RTX 3090, I produced a peer-review-accepted result in a direction the field talks about but rarely operationalizes. Everything is public and reimplementable. I currently paint game key-visuals full-time; funding replaces ~6 months of that income so this becomes my full-time work. Disclosure: code, experiments, and drafting made substantial use of AI systems (Claude and others); I am sole author and take full responsibility. AI contributors are credited in acknowledgements. DOIs: zenodo.20783830 · 19971267 · 19960838 · 19612585
Most likely failure mode: the individuation signal does not survive statistical scaling (n≥40) — the n=8 directional result turns out to be underpowered noise. Outcome if so: a documented, public negative result with full replication scripts, which still informs the model-identity discussion (knowing the signal is fragile is itself useful). Secondary risk: the transformer-bridge extension (months 4–6) yields a negative result — this is explicitly exploratory and budgeted as such. The core deliverable (a re-implementable protocol + powered data) survives both failure modes. Funds are spent linearly; no large irreversible commitments.
None. The project has been entirely self-funded to date (personal hardware, personal income from full-time illustration work). This same proposal is currently under evaluation at the EA Long-Term Future Fund (no decision yet).