Nemotron Rl Instruction Following Adversarial V1

A dataset of adversarial prompts designed to explicitly conflict with an AI model's standard training instincts, such as writing code without comments or refusing helpfulness norms. It focuses on 8 distinct 'anti-convention' patterns and was created by NVIDIA, with a last recorded update on March 11, 2026. The dataset uses a targeted 'model breaking' methodology to generate candidate responses for testing constraint difficulty.

Use Cases

• Testing model robustness based on adversarial prompts that conflict with standard training instincts
• Evaluating instruction-following failure modes based on 8 distinct 'anti-convention' patterns
• Benchmarking AI safety and alignment based on responses to prompts designed to break helpfulness norms
• Analyzing model behavior under negative constraints based on a 'model breaking' methodology

Strengths

• Focuses on 8 distinct 'anti-convention' adversarial patterns
• Created by NVIDIA, a major AI research organization
• Last updated March 11, 2026

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment
• Description metadata is limited; actual data quality requires manual inspection after download

Provenance

Source

NVIDIA

Collection Method

Generated using a targeted 'model breaking' methodology with candidate responses from models like Nemotron-Nano-V2 or Qwen3-235B-A22B-Thinking-2507.

Time Range

Last updated March 11, 2026.

Freshness

Last updated 2026-03-11 04:34:12; freshness should be verified

Geography

null