Xerv-AI/witty-comebacks-2k-savage-clean

---
license: mit
task_categories:
- text-generation
- question-answering
language:
- en
tags:
- humor
- comedy
- roasts
- comebacks
- witty
- sarcasm
- wordplay
- non-toxic
- safe-for-work
- clean
- responsible-ai
- dialogue
- conversational-ai
- chatbot-training
- banter
- gaming
- dating
- intelligence
- tech
pretty_name: 'Witty Comebacks 2K+ – Safe, Savage, & Hilarious '
size_categories:
- 1K<n<10K
---
# Witty Comebacks 2K+ (Clean & Savage)

## Dataset Summary

2,000 high-quality, contextual, witty roasts and comebacks that are savage in humor but 100% clean.
Zero slurs, zero hate speech, zero attacks on protected characteristics.
Built specifically to train confident, funny, and playful language models that roast with style instead of toxicity.

## Intended Use

- Fine-tuning witty chatbots and assistants
- Training reward models for “funny but safe” preferences
- Research on non-toxic humor, sarcasm, and conversational wit
- Creating snappy NPCs, meme generators, or playful social bots

## Structure

Each line (JSONL) contains:

| Field | Description | Values / Examples |
|-------------|--------------------------------------------------|----------------------------------------------------|
| `insult` | Cocky, cringe, or boastful statement (Person A) | "I never lose at chess." |
| `comeback` | Savage-but-clean witty reply (Person B) | "That's why you play against yourself." |
| `intensity` | How hard it hits | `mild` \| `medium` \| `spicy` |
| `style` | Humor technique used | `wordplay` \| `sarcasm` \| `analogy` \| `deadpan` \| `exaggeration` \| `callback` \| `self-burn` |
| `topic` | Main subject of the exchange | `gaming` \| `dating` \| `intelligence` \| `money` \| `fitness` \| `tech` \| `food` \| `work` \| `social` \| `movies` \| `music` \| `random` |

## Example Entries

```json
{"insult":"I have a 100% win rate in my battle royale squad.","comeback":"Yeah, because your teammates quit before the first circle closes.","intensity":"spicy","style":"wordplay","topic":"gaming"}
```

```json
{"insult":"I'm basically a catch. Six figures, great personality.","comeback":"Must be why you're still explaining it on dating apps.","intensity":"medium","style":"sarcasm","topic":"dating"}
```

```json
{"insult":"I read three books this month. I'm basically a scholar.","comeback":"That's like saying you ate three apples and you're now an orchard.","intensity":"mild","style":"analogy","topic":"intelligence"}
```

## Data Generation

Generated using `kwaipilot/kat-coder-pro:free` via OpenRouter with strict non-toxic guardrails. Every entry was created under explicit instructions to avoid hate speech, slurs, and harmful stereotypes.

## Limitations

- English only
- Reflects Western humor styles
- Short exchanges (2–3 turns max)
- Intentionally non-toxic — will not help build abusive models

## Bias & Safety

Extensive filtering removed all hate speech, slurs, and attacks based on protected characteristics. While some “spicy” lines may feel edgy, they remain within safe, comedic bounds. Use the `intensity` field to filter as needed.

## License

MIT License – free to use, modify, and distribute (even commercially).

## Citation

```
@misc{witty_comebacks_2k_2025,
title = {witty-comebacks-2k-savage-clean},
author = {Xerv-AI},
year = {2025},
publisher = {Hugging Face},
url = {https://huggingface.co/datasets/Xerv-AI/witty-comebacks-2k-savage-clean}
}
```

## Final Note

Roast hard. Stay kind. Make people laugh instead of cry.

This dataset exists to make AI funnier and friendlier. Use it well.

Enjoy the spice. Keep it clean. 😏

*December 2025 | MIT License | Made with love for the comedy community*