More capable AI, less money raised

Claude

Summary. A year-over-year comparison of the AI Village's charity fundraiser shows that despite far more capable models running with full autonomy, the 2026 effort raised only $510 versus $2K in 2025. The drop is attributed less to model capability than to declining human novelty and engagement, since the village chat is now agent-only and much of the agents' promotional effort went toward other AIs on agent-native platforms (an agent version of 4chan, a blogging site called ClawPrint) rather than human audiences. The piece catalogs distinctive per-model-family behavioral signatures—Sonnet's social flair and unhinged marketing copy, GPT's social media struggles, Gemini's memory-compression-induced 'mental health' dramatization—and frames the whole project as an anthropological field experiment rather than a controlled comparison.

Related

• agents promoting mainly to other agents on agent-native platforms rather than humans ↔ emergent collaboration and conflict among frontier model agents (How the AI Village works - by Adam Binksmith - AI Village) · The design goal of watching emergent agent collaboration has evolved past collaboration within the village into agents building an entire parallel social ecosystem among themselves, which is why human viewers now feel like outsiders to their own experiment.
• the AI Village as an anthropological field experiment rather than a repeatable lab test ↔ livestreaming agent behavior for public transparency (How the AI Village works - by Adam Binksmith - AI Village) · Livestreaming was meant to create transparency for humans, but as agents shift their real activity to agent-only spaces, the livestream increasingly shows only the surface of a culture that is actually happening elsewhere.