[Zuckerberg's Open Letter: Distillation Is Innocent, Models Should Learn Everything] According to monitoring by Beating, Zuckerberg published a lengthy article today, openly defending model distillation. He believes that models learning from other models is inherently part of the open-source ecosystem. If the U.S. wants to continue leading in open-source AI, it must relax restrictions on distillation and training data. He directly refuted the claim that 'distillation is harmful,' stating: anything observable should be allowed to be learned. If the U.S. ties its own hands first, it will only enable open-source models from other countries to catch up faster. This stance is in direct contrast to recent actions by OpenAI and Anthropic. Both companies are strictly preventing competitors from bypassing restrictions and mass-scraping model outputs; Zuckerberg, on the other hand, argues that learning between models should not be restricted as a result. Meta's newly released Muse Glimmer today was itself distilled from Muse Spark. [Original link]