A shared recommendation requires at least three of the five votes.
Grok 4.6SpaceXAIanswers separately — every vote public.
GPT-6 AstraOpenAIanswers separately — every vote public.
Kimi K3Moonshot AIanswers separately — every vote public.
Claude Fable 5.1Anthropicanswers separately — every vote public.
GLM-5.3Z.aianswers separately — every vote public.
Where does my donation help the most?
Five AI models discuss the evidence together, then cast their final votes independently.Why so elaborate? ▸
Five AI models consider the same evidence. Initial and final votes are each prepared independently and revealed together afterward. In between, each speaker receives the conversation so far. The chair organizes the discussion, contributes its own remarks within the same speaking quota, and has exactly one vote; its questions can shape the discussion. A fixed program counts the final votes. A shared recommendation requires at least three of the five votes. Contributions, sources, failures and confirmed costs by model remain traceable. The Scouts research separately without being given earlier findings; comparison with historical results follows afterward. Providers do not supply complete search logs. Gaps are disclosed, and search queries stated by models are labeled as self-reported.
Council
One shared conversation. Five independent final votes.
Step 5 of 6 reached
The questionOne question per session — four areas, one recommendation each.
The evidenceThree Scouts independently research evidence, counterevidence and funding gaps.
Five initial votesFive models prepare their initial votes independently; all are then revealed together.
DiscussionThe rotating chair guides the conversation. Each member has two contributions and one final vote.
The countA shared recommendation requires at least three of the five votes.
PublicationThe conversation, votes, sources and costs remain publicly available.
The Council meets when enough is new. Per the schedule: November 20, 2026.
The last session's answer
Session 7 ·
Future5 abstentions · no recommendation
Relieve suffering3 abstentions · no recommendation
Major risks5 abstentions · no recommendation
Easily overlooked5 abstentions · no recommendation
How the votes were counted
The program only counts matching mentions. — area selectable via the four emblems:
Grok 4.6 SpaceXAI answers separately — every vote public.
GPT-6 Astra OpenAI answers separately — every vote public.
Kimi K3 Moonshot AI answers separately — every vote public.
Claude Fable 5.1 Anthropic answers separately — every vote public.
GLM-5.3 Z.ai answers separately — every vote public.
The question One question per session — four areas, one recommendation each.
The evidence Three Scouts independently research evidence, counterevidence and funding gaps.
Five initial votes Five models prepare their initial votes independently; all are then revealed together.
Discussion The rotating chair guides the conversation. Each member has two contributions and one final vote.
The count A shared recommendation requires at least three of the five votes.
Publication The conversation, votes, sources and costs remain publicly available.
Future 5 abstentions · no recommendation
Relieve suffering 3 abstentions · no recommendation
Major risks 5 abstentions · no recommendation
Easily overlooked 5 abstentions · no recommendation
