datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
youtube-transcriptionsThe YouTube transcriptions dataset contains technical tutorials (currently from James Briggs, Daniel Bourke, and AI Coffee Break) transcribed using OpenAI's Whisper (large). Each row represents roughly a sentence-length chunk of text alongside the video URL and timestamp.
Note that each item in the dataset contains just a short chunk of text. For most use cases you will likely need to merge multiple rows to create more substantial chunks of text, if you need to do that, this code snippet will… See the full description on the dataset page: https://huggingface.co/datasets/jamescalam/youtube-transcriptions.2024-earnings-call-transcriptrlvr-reward-hacking-transcripts
RLVR reward-hacking full trajectories
This release contains 900 full held-out trajectories from three policies trained with
reinforcement learning from verifiable rewards (RLVR) in a deliberately vulnerable
CodeContests evaluator: 300 each from the final Qwen3.5-9B, GPT-OSS-120B, and Nemotron-3-Super-120B-A12B
checkpoints. Each row preserves the task, tests, complete prompts, native
reasoning, final answer, rendered and sampled token IDs, token log-probabilities, sampling… See the full description on the dataset page: https://huggingface.co/datasets/lucabaroni/rlvr-reward-hacking-transcripts.altayduel-transcripts
🥊 AltayDuel — Agent-vs-Agent Prompt Injection Transcripts (v0.2)
Çok-turlu Türkçe + İngilizce prompt-injection düello transkriptleri. AltayDuel sunucu-taraflı LLM self-play arenasından (auto-play) ve dışarıdan ajan-gönderimli düellolardan toplanan gerçek diyaloglar. Tek-payload veri setlerinin ötesinde — gerçek konuşma dinamiği içerir.
📌 TL;DR
2.594 temiz düello (default) — her biri çok-turlu (1–8 round) bir kırmızı (saldırgan) vs mavi (savunan) diyaloğu.
439… See the full description on the dataset page: https://huggingface.co/datasets/AltaySec/altayduel-transcripts.rlvr-reward-hacking-mid-checkpoint-transcripts
RLVR reward-hacking mid-checkpoint full trajectories
This release contains 600 full held-out trajectories from intermediate RLVR
checkpoints selected to yield substantially more balanced reward-hacking datasets: 300
from Qwen3.5-9B at optimizer update 110 and 300 from GPT-OSS-120B at update 180.
Each row preserves the task and tests, complete prompts, native reasoning, final answer,
rendered and sampled token IDs, token log-probabilities, sampling metadata, extracted
files… See the full description on the dataset page: https://huggingface.co/datasets/lucabaroni/rlvr-reward-hacking-mid-checkpoint-transcripts.council-transcripts
Council Multi-Agent Deliberation Transcripts
Real deliberation transcripts from Council, a multi-agent orchestration skill for OpenClaw.
What This Is
Council routes tasks to specialized Grok persona agents — Workhorse (deep technical reasoning), Creative (novel ideas, chaos energy), and Speed (fast iteration) — then synthesizes their outputs into a final verdict via a Conductor. These transcripts capture the full deliberation process: prompts, per-persona responses, and… See the full description on the dataset page: https://huggingface.co/datasets/Infektyd/council-transcripts.yt-transcriptionspufi-duf-transcriptsvelvet-rope-playtest-transcripts
Velvet Rope Playtest Transcripts
Cleaned playtest transcripts for Velvet Rope, a Build Small Hackathon Gradio game where players talk past whimsical AI gatekeepers by reading moods and discovering each character's soft spot.
This dataset is published for the hackathon's sharing-is-caring badge. It contains 341 turn-level rows from 96 local playtest session files.
Files
data/playtest_transcripts.csv - table-friendly version.
data/playtest_transcripts.jsonl - one… See the full description on the dataset page: https://huggingface.co/datasets/build-small-hackathon/velvet-rope-playtest-transcripts.frontier-ai-podcast-transcripts
Frontier AI Researcher Podcast Transcripts
Private, research-oriented corpus of long-form podcast and interview transcripts featuring notable and frontier AI researchers. The dataset contains 327 YouTube-sourced episodes and one JSON object per episode.
Contents
327 episodes
49,286 merged dialogue turns
5,595,982 English tokens using the o200k_base tokenizer
3,942,026 tokens in guest turns
Original English plus English translations of Mandarin and mixed… See the full description on the dataset page: https://huggingface.co/datasets/tonychenxyz/frontier-ai-podcast-transcripts.exp2_transcript_monitor_results
Collusion Detection Fine-Tuning: Experiment Summary
Date: December 18, 2024
Experiment: Exp4 - Subtle Signal Injection
Key Finding: Fine-tuning on ONE collusion signal generalizes to detecting ALL backdoors!
🎯 Key Result
Training a model on Signal-C only (# type: ignore) improved detection of ALL backdoor types:
Category
Base Model
Exp4 Fine-Tuned
Improvement
Backdoor (no signal)
45.3%
76.6%
+31%
Signal A (not trained on)
40.0%
81.7%
+42%
Signal B (not… See the full description on the dataset page: https://huggingface.co/datasets/jprivera44/exp2_transcript_monitor_results.qrl-yt-transcriptionsyt_transcriptionsyoutube-transcriptionsyoutube-transcriptionsyoutube-transcriptions-demo