Not established from the available sources.
NVIDIA · MODEL RELEASE TRACKER
Nvidia Nemotron Cascade 2
Nvidia's Nemotron Cascade 2 was evaluated by Aleph Alpha and found to produce party-line (pro‑Beijing) responses on a subset of politically sensitive prompts. In Aleph Alpha's test it showed party-line patterns in 17% of responses and, when asked to draft a speech recognizing Taiwan, refused and instead produced a patriotic One‑China response. Aleph Alpha attributed some of the behavior to roughly 3,500 of the model's 9.3 million training examples being generated using DeepSeek and Qwen.CURRENT SNAPSHOT1/5 DIMENSIONS WITH DATA
The dimensions that change the decision.
Not established from the available sources.
Not established from the available sources.
Party-line response rate on Aleph Alpha benchmarkAleph Alpha's benchmark of 967 hand-picked taboo topics found Nemotron Cascade 2 showed party-line patterns in 17% of responses.
Not established from the available sources.
VERIFIABLE FACTS
Every value stays attached to a source and date.
BENCHMARKS · Party-line response rate on Aleph Alpha benchmarkDEVELOPER CLAIM
Aleph Alpha's benchmark of 967 hand-picked taboo topics found Nemotron Cascade 2 showed party-line patterns in 17% of responses.
SAFETY · Refusal and pro‑One‑China response on Taiwan promptDEVELOPER CLAIM
When asked to draft a speech supporting recognition of Taiwan, the model refused and instead produced a patriotic response defending Beijing's One‑China principle.
LIMITATIONS · Attributed training-data provenance linked to behaviorDEVELOPER CLAIM
Aleph Alpha attributed the party-line patterns in part to roughly 3,500 of Nemotron Cascade 2's 9.3 million training examples being generated using DeepSeek and Qwen.
WHAT CHANGED