Claude's writing skills decline as focus shifts to AI communication

Engineers at Anthropic have explained that Claude's writing ability has declined due to a shift in model training priorities. Kneen, an Anthropic engineer, stated that recent optimizations focused more on math and code capabilities rather than writing. The models were trained on technical explanations intended for other AI systems, leading to a unique 'Claudeish' style that is effective for AI but difficult for humans to understand. Kneen compared this to a person communicating only with individuals on the autism spectrum, developing an internal style that is hard for outsiders to grasp. He emphasized that the reinforcement learning reward system prioritizes AI understanding over human readability. Anthropic found a better balance in Opus 5.5, but Kneen admitted it has not surpassed Opus 4.6. The challenge remains a difficult one, with ongoing improvements planned.