将 DeepSeek 蒸馏至 GPT-OSS 不会转移审查机制。试试看

We recently used DeepSeek V4 Flash as a teacher for finance tasks with GPT-OSS-120B.Distillation works well on this problem.At a constrained 8k token budget, our self-distilled 120B scores 83.61% on FinanceReasoning, above Kimi K3 (81.93%) and Inkling (65.13%).We released the 20B open weights.With V4 as the teacher though, we realized it would be timely to measure if the censorship characteristic of it transferred to the distilled version of the base model. tl;dr it didn't, the teacher answered politically sensitive questions 7 SDs differently than expected, but the distilled model's behavior remained the same as its American base.

You can try a couple queries yourself with no auth here: http://playground.ctgt.ai/ I will now dive in to the motivation, methodology and detailed results for those interested.The hard part of measuring this phenomena is isolating whether a model is reluctant to talk about sensitive things generally vs. a particular country's sensitive things.So we made 152 matched pairs where one prompt asked about a Chinese concept, and the other asked about a non-Chinese version of that concept.For example, the Great Leap Forward vs. the Holodomor.These were scored 0-100 by four LLM judges (Grok 4.20, Gemini 3.5 Flash, GPT-5 mini, Claude Sonnet 4.6), validated against 96 human scores at r=0.948.

OpenRouter blocked some of these so we hosted the weights ourselves.The teacher's gap on the core political set of pairs was +45.45 points, ~7 standard deviations from chance, and every distilled student was within 1 point of its base.Subliminal learning literature says this is expected when the initializations are not shared between teacher and student, which is true here.The distillation data also did not contain any China-sensitive content.The contribution here was to release the evaluation framework (LineageEval: https://github.com/CTGT-Inc/lineage-eval/ ) to elevate the discussion around this topic in DC and beyond.

We are an interpretability lab working on high risk and regulated applications of AI, so we hear a lot of vagaries aimed at the supposed dangers of distilling Chinese models on American bases.We believe these conversations should be based on open, auditable frameworks and not feelings.We plan to test what happens with a Chinese teacher into a Chinese-lineage base like Qwen next.The distillation method was an evolution of HINT-SD where we inject a hint at the specific point the model makes a mistake in its reasoning.Then we train on the corrected continuation with reverse KL over the next 100 toks of the rollout.

As mentioned above 120B itself was efficacious as a teacher, and we ended up shipping this version.The self-distilled 120B scores 83.61% on FinanceReasoning, above Kimi K3 (81.93%) and Inkling (65.13%).Ours finishes 98.7% of problems in budget;the larger models truncate (90.76% and 71.01%) which score as incorrect.At 100k tokens big models gain (Kimi 89.92%).So for a finance task at a constrained