ByteDance bans rival model distillation amid US-China AI tensions

The distillation ban reveals the mounting pressures on Chinese AI makers caught between Beijing's demand for efficient, domestically-built models and the reality that smaller competitors have begun outpacing ByteDance's own systems. Model distillation—the practice of compressing knowledge from larger AI models into smaller, faster versions—has emerged as a lightning rod in US-China AI competition, with Washington viewing it as a potential workaround to export controls. ByteDance's 2023 decision to forbid the practice internally suggests the company anticipated the geopolitical risks years before distillation became a flashpoint for sanction threats. The policy's timing, the newsletter notes, predates the period when "avoiding US retaliation" became an explicit commercial priority, hinting that ByteDance may have had other competitive or strategic reasons for the restriction. That the rule has sparked intense internal debate points to a deeper tension: while ByteDance aims to comply with implicit US pressure, the company's AI ambitions are being outpaced by smaller, faster-moving domestic rivals, even without distillation. US models currently deliver better value, but the gap is narrowing as Chinese labs innovate under resource constraints.