Interactionindex: Synthetic Preference Data for Coding and Safety Tasks

A synthetic preference dataset created by 8F-ai and last updated in March 2026. It is organized into four subsets focused on coding tasks, safety-sensitive refusals, honesty checks, and everyday assistant behavior. The dataset is designed for preference modeling, dataset tooling, and RLHF-style experimentation.

Use Cases

• Training preference models based on synthetic human feedback for coding tasks.
• Experimenting with reinforcement learning from human feedback (RLHF) using the provided safety-sensitive refusal data.
• Developing dataset tooling for handling structured preference data across multiple behavioral subsets.
• Benchmarking AI assistant honesty and refusal behavior in coding and safety contexts.

Strengths

• Dataset is structured into four distinct subsets for focused analysis.
• Last updated on 2026-03-27, indicating recent maintenance.

Limitations

• Row count, file formats, and column-level documentation are unknown.
• Data is synthetic, which may limit realism compared to real human feedback.
• License information is unspecified, requiring verification before use.

Provenance

Source

8F-ai via Hugging Face Datasets.

Collection Method

Synthetically generated.

Time Range

Synthetic data; no temporal coverage specified.

Freshness

Last updated 2026-03-27 17:54:27.

Geography

Synthetic data; no geographic coverage specified.