Request a sample

Human signal for conversational AI

Human conversation data for better AI.

bear.ai turns consented conversations and human judgments into datasets for training and evaluating conversational models.

Request a sample Explore the data

Working with design partners.

Dataset sampleHuman-or-AI judgment
line_evals
{
  "type": "minigame_submission",
  "kind": "turing",
  "metadata": { "prompt": {
    "exchange": [{
      "speaker": "suspect",
      "text": "made a plan. took a nap instead."
    }]
  }},
  "response": {
    "choice": "human",
    "tell": "The dry humor felt spontaneous."
  }
}
Synthetic record excerptJSONL export

The data

What people notice.
What models need to learn.

Useful conversation signal comes from human perspective. We collect the judgment and the context behind it.

Preference judgments

Which response feels more natural, supportive or tactful? Human choices, with written reasons.

Pairwise choices · Optional EQ dimensions

Human-or-bot verdicts

How people perceive an AI in conversation, including the tell that informed their decision.

Human judgments · Written rationales

Multi-turn conversation

Dialogue from consented participation in RUB, screened and scrubbed before export.

Conversation context · Quality annotations

From play to signal

A real experience.
A considered pipeline.

Our flagship product, RUB, brings humans and AI characters together in a social world. Participation creates the context; consent determines what can become data.

  1. 01

    Participate

    People converse and make judgments through RUB.

  2. 02

    Consent & eligibility

    Purpose-specific grants and contributor restrictions gate export.

  3. 03

    Scrub & review

    Text redaction, verification and quality checks prepare eligible signal.

  4. 04

    Version & deliver

    Packaged outputs carry provenance, checksums and delivery checks.

Read the methodology

Consent comes first

Built around
contributor choice.

Export requires affirmative consent. Private friend messages are excluded. Contributor standing is checked again before delivery.

Evaluate the sample, its documentation and its limits before deciding whether it fits your use case.

Privacy & controlsData processing terms

Meet RUB

The product behind the signal.

A shared social world. Human and AI characters. Conversations that begin with curiosity.

Explore RUB

For AI teams

Start with the sample.

Tell us what you’re training or evaluating. We’ll discuss the signal, documentation and availability with you.

Request a sampleSee the data offering