New todaydeepseek-v4.1POST
Report: DeepSeek and Moonshot distilled from GPT-6 Astra
An English recap of The Information's report that V4.1-Flash 1017 and Kimi K3.1 Pro trained heavily on Astra outputs, with Astra becoming the main target after Anthropic tightened defenses; no company response yet.
RC | Macro Sift
@themacrosiftArchived 48 min agoOriginally posted 2026.10.29
RC | Macro Sift @themacrosiftOriginally posted 2026.10.29
🚨 China's top AI labs reportedly found a new teacher: OpenAI's GPT-6 Astra. The Information reports DeepSeek and Moonshot leaned heavily on Astra's outputs to train their newest models, DeepSeek V4.1-Flash 1017 and Kimi K3.1 Pro. Astra reportedly became the main distillation target after Anthropic tightened its defenses. No public response from the companies yet. If it holds, frontier labs now have to guard their outputs as closely as their weights.
x.com/themacrosift/status/2115652546308039156Archived 2026.10.29 12:04