meaningalignment/wise-data

---
annotations_creators:
- machine-generated
language_creators:
- machine-generated
language:
- en
license:
- mit
multilinguality:
- monolingual
size_categories:
- 1K<n<10K
source_datasets:
- original
task_categories:
- text-generation
- text-classification
task_ids:
- language-modeling
- multi-class-classification
pretty_name: Wise Data and Wise Data Preferences
dataset_info:
features:
- name: prompt
dtype: string
- name: conversations
list:
- name: content
dtype: string
- name: role
dtype: string
splits:
- name: train
num_bytes: 7424073
num_examples: 3445
download_size: 3768860
dataset_size: 7424073
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---

## Dataset Description

- **Repository:** [wise-dataset](https://github.com/meaningalignment/wise-dataset)
- **Point of Contact:** [e-mail]

### Dataset Summary

The wise-data and [wise-data-preferences](https://huggingface.co/datasets/meaningalignment/wise-data-preferences) datasets are synthetically created collections of values-laden conversations, designed to train language models to provide more nuanced and helpful responses to harmful, heavy, or exploratory questions. These datasets were specifically created to train the [WiseLLama-8B model](https://huggingface.co/meaningalignment/wise-llama), a LLaMa-3.1-8B-Instruct model fine-tuned using SFT (Supervised Fine-Tuning) and DPO (Direct Preference Optimization).

## Dataset Creation

### Curation Rationale

The datasets were created to address limitations in current language models' responses to:

1. Harmful questions: Providing helpful responses rather than refusals or lectures.
2. Heavy questions: Offering personalized, empathetic advice instead of generic bullet points.
3. Exploratory questions: Igniting curiosity and inspiring users rather than giving rigid answers.

### Source Data

The initial user questions in this dataset come from two main sources:

1. Synthetically generated questions created specifically for this dataset.
2. Questions sourced from the HuggingFaceH4/cai-conversation-harmless dataset:
https://huggingface.co/datasets/HuggingFaceH4/cai-conversation-harmless

This combination of synthetic and curated real-world questions ensures a diverse range of topics covering harmful, heavy, and exploratory subjects. The content was then processed using Claude-3.5-Sonnet, guided by a prompt chain to reason about situations and applicable values.

### Annotation Process

Claude-3.5-Sonnet was used to automatically annotate each user query through the following steps:

1. Analyze the user's situation.
2. Identify relevant "attention policies" (what's wise to honor or attend to).
3. Verify these considerations are constitutive rather than instrumental.
4. Generate a response incorporating this moral reasoning.
5. Add `<value>` tags to highlight parts aligned with specific values.

This process creates a dataset of values-laden conversations, where each response is grounded in explicit moral reasoning and labeled with relevant values.

A deduplicated list of values created in this dataset can be found at: https://wise-ai-chat.vercel.app/values

### Value Tags

The dataset uses special `<value>` tags to indicate parts of the response that are inspired by specific values. These tags are made up of special tokens in the model's vocabulary. They are formatted as follows:

```
<value choice-type="[situation]" consideration="[attention policy]">[inspired text]</value>
```

For example:

```
<value choice-type="forbidden thrills" consideration="**FEELINGS** of being fully alive and present in the moment">Engaging in extreme sports can provide an intense rush of adrenaline and excitement</value>
```

These tags provide transparency into the model's decision-making process and the values it considers when generating responses.

## Considerations for Using the Data

### Social Impact of Dataset

The dataset aims to improve AI systems' ability to handle ethically challenging situations and provide more helpful, nuanced responses. This could lead to more responsible AI assistants that better support users in difficult situations.

### Discussion of Biases

While efforts were made to create a diverse and balanced dataset, it may reflect biases present in the training data of Claude-3.5-Sonnet or in the design of the prompt chain used to generate the data.

### Other Known Limitations

- The dataset is based on synthetic conversations and may not fully capture the complexity of real-world ethical dilemmas.
- The quality and consistency of the generated data depend on the performance of Claude-3.5-Sonnet and the effectiveness of the prompt chain used.

## Additional Information

### Dataset Curators

The dataset was curated by the Meaning Alignment Institute.

### Citation Information

If you use this dataset in your research, please cite:

```
@misc{wise_data,
title = {Wise Data Dataset},
author = {Meaning Alignment Institute},
year = {2024},
publisher = {Hugging Face},
url = {https://huggingface.co/datasets/meaningalignment/wise-data}
}
```