Firehose

Filtered to Hacker News, tagged “Censorship” · clear filters

All PeopleCompaniesPapersPodcastsHacker News

Browse by tag

30 JUL 2026 · Hacker News · 165 pts · 72 comments ↗

Researchers found that distilling a Chinese frontier model (DeepSeek V4 Flash) into a self-distilled model (CTGT 120B) does not transfer censorship, despite training on the same outputs. The self-distilled model outperformed a base model (GPT-OSS-120B) on finance-related tasks, with similar performance to a more advanced Chinese teacher model (DeepSeek V4 Flash). AI summary