Sample data:

Direct Preference Optimization Formatted: subjective_response_DPO_sample.json

Customizable Format for RLHF: subjective_response_RLHF_data.json

What does the company do?

Us Folks is building a civic-tech platform that captures nuanced political values and tests how AGI might advocate for citizens. While our MVP for gathering novel subjective preference data is still a few weeks away, we currently have specialized pipelines to process and structure historical public-opinion data into anonymized, respondent-level datasets suitable for RLHF training.

Data collection and scale

Us Folks processes respondent-level microdata from political, public-sentiment, or consumer surveys into RLHF datasets. We also transform crosstabs-only data into statistically representative synthetic microdata, unlocking huge quantities of suitable data—especially election exit polls and leading-firm political polling, recent and historical.

For each response we provide:

• Survey metadata (e.g., date, collection mode, polling firm)
• Respondent profile & demographics (age, gender, race, education, region, etc.)
• Prompt (the subjective question)
• Response (the respondent’s answer)

We generate structured (profile → prompt → subjective response) tuples. We can create millions of tuples per week, drawing on multi-decade polling archives with open data-use licenses.

Unique value for customers

• Large-scale, subjective political & social data under-represented in current RLHF corpora
• Enables personalized, subjective LLM use-cases that require insight into moral values, governance preferences, and human judgment at scale
• Critical for modeling personalized decision-making needed for aligned AGI
• Pipeline will soon include brand-new subjective datasets from the Us Folks MVP platform