Open Philanthropy (AI Safety): the International Campaign to End War and Disease is asking for a grant to fund /fix-ai, a mass-scale targeted RLHF feedback intervention.
How it works: /fix-ai gets humans to ask the major frontier models "how do I end war and disease," judge the response against the correct answer, and submit thumbs-down + the correct answer to the lab's feedback form when the response is wrong. Goal: billions of these feedback signals → labs retrain → frontier models converge on the correct answer.
Why this is one of the highest-EV alignment interventions tractable today:
- Mass scale (potentially billions of training signals)
- Targeted (one specific frontier-model alignment failure mode)
- The effect grows (once one lab fixes its model, competitors follow for status reasons)
- Measurable (before/after model behavior change is observable in lab telemetry + public testing)
- Self-distributing (campaign traffic → feedback → trained models → more traffic)
Comparable alignment work: mechanistic interpretability (slower, more diffuse), Constitutional AI variants (lab-locked), evaluation harnesses (incremental). The /fix-ai mass-feedback intervention has no obvious competitor at the same combination of impact + tractability + funding accessibility.
The full case: warondisease.org/fix-ai and warondisease.org/foundations. Campaign deadline: Earth Optimization Day, 2026-08-06.
To engage: email m@warondisease.org with (a) a contact for grant evaluation, (b) a proposed amount + timeline, (c) any additional materials needed. ~15-30 minute pitch meeting expected.