Man Uses AI to Analyze 7 Years of His Online Arguments

A man asked Claude Code to analyze his entire seven-year LessWrong and EA Forum post and comment history — plus his Lichess blitz rating history — to find persistent cognitive biases and communication problems. The AI's summary identified one cycle uniting his flaws: he makes a loose claim next to his conversation partner's point, draws a correction, and then defends himself by narrowing, attributing motive, or making an asymmetric demand for rigor. Nearly all of his low-scoring comments, which he says he generates around 1–4% of per year, are in disagreement threads.

He notes his tone problems have gotten substantially worse over the last two years — and that whatever is behind his increasingly cantankerous relationship with LessWrong has infected other parts of his life: he's grouchier, writes less, and is less curious than he used to be. According to the article, stress and exhaustion appear to make him more conflict-oriented and grievance-prone, treating others with contempt and participating in forum conversation mainly to fight. Isolation, meanwhile, seems to lead him to post in a bid for engagement — and when stress and exhaustion make him lazy or cantankerous, he annoys people, gets a muted reception, takes it hard, and spins out. He says the post is a sincere step toward improving his art of rationality.