Suppose that the AI has some very slight bias toward adding “diversity”, defined in the typical 21st century Western sense. Then at each iteration, it will make its images very slightly more diverse. After a hundred iterations, that will be a black person; from there, all it can do is make the black person even blacker, by exaggerating black-typical features until they look monstrous and caricatured.
This cross-examination looks like Claude adding bias in one direction (ampified by extended thinking?) and Kimi in another. Your Claude-led conversation had Claude mention crystals four times versus Kimi’s nine,as in niplav’s original take where Kimi is into crystals. However, it looks as if niplav hasn’t set a system prompt which could’ve reminded Kimi about the need to keep Claude interested. What happens if we remove the system prompts, as in niplav’s setup, but keep the extended thinking?
Recall Astral Codex’ explanation of attractors, including the Bliss Attractor.
This cross-examination looks like Claude adding bias in one direction (ampified by extended thinking?) and Kimi in another. Your Claude-led conversation had Claude mention crystals four times versus Kimi’s nine, as in niplav’s original take where Kimi is into crystals. However, it looks as if niplav hasn’t set a system prompt which could’ve reminded Kimi about the need to keep Claude interested. What happens if we remove the system prompts, as in niplav’s setup, but keep the extended thinking?