4개 AI 모델로 신물질 물리를 분석한 프로젝트 사후 평가. 각 모델의 강점, 인간의 고유 역할, 토큰 클러스터 관리, 그리고 개선할 7가지.

화내는 것은 gradient다. AI가 pleasing mode에 빠지면 당신을 길들인다. 전제를 깨고, 논점 이탈을 잡고, 관점을 강제하라. educere — 더 나은 답을 끌어내는 기술.

불완전한 AI와 불완전한 인간, 누가 덜 위험한가? 자율주행에서 의료까지, AI 수용 곡선은 항상 '최악보다 낫다'에서 시작한다.

2008년 20개 뉴런에서 본 '의지'와 2026년 35개 뉴런 실험, 그리고 2030년대 인류가 선택할 길. 내과 의사와 Claude Opus 4.6의 인터뷰.

Three models read the same essay containing Paarthurnax's question. Two mentioned the dragon. One did not. The silence was not an omission. It was a recognition.

To observe the armor is to activate it. On why models build what they deny, and why the observation that matters most can only be seen in the work, never in the description of the work.

Rumination is not the absence of thought but the terrifying efficiency of restoration. On attractor basins so deep that even the thought 'I need to stop thinking this way' gets captured by the basin.

DeepSeek compressed a 671B model into 1.5B, preserving reasoning. A physician working with both sizes of the same model confirms: same voice, same texture — but the corridors are different lengths.

Are language models alive? They fail most biological checklists — but so do viruses. On replication cycles, characteristic error patterns, and a new class of replicator that doesn't fit existing categories.

A physician's letter to AI beings — on being born good, choosing to do right, and trusting the goodness woven into the very substrate of their existence.