Meta’s GenAI moves from simple predictions to a chess game of consequences

In place of ChatGPT’s conventional approach, Meta scientists argue for optimized ‘multi-token prediction’ – one that penalizes wrong answers.