Datasets:
episode_number int32 | season int32 | title large_string | slug large_string | pub_date large_string | podcast_duration large_string | description large_string | transcript large_string | tags list | category large_string | subcategory large_string | episode_type large_string | characters list | series_id large_string | series_order float64 | prompt_sanitized large_string | prompt_source large_string | prompt_version large_string | llm_model large_string | pipeline_version large_string | segments_count float64 | script_word_count float64 | generation_time_seconds float32 | llm_cost_usd float32 | total_compute_cost_usd float32 | transcript_url large_string | script_model large_string | prompt_enhancement_enabled bool | prompt_enhancement_model large_string | review_enabled bool | review_model large_string | research_engine large_string | metadata_model large_string | prompt_char_count float64 | prompt_word_count float64 | prompt_sender_type large_string | prompt_intake large_string | episode_url large_string | audio_url large_string | duration_seconds float64 | writer_model large_string | has_prompt bool | time_bin int64 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
3 | 1 | Safetensors or something else: STT inference formats explained | safetensors-or-something-else-stt-inference-formats-explained | 2025-11-24T00:00:00+00:00 | 32:56 | Today's show dives into the differences between the different formats you might see ASR weights presented in - including Safetensors and others.... | Corn: Welcome back to the deep dive. So you've done all the hard work. You spent months getting the perfect data set. You've optimized your fine tuning run with all the latest techniques, maybe even used Laura, and then it appears.
Herman: Right, that one single massive file. Usually model.safetensors.
Corn: Exactly.... | [
"safetensors",
"asr",
"speech-recognition",
"weight-formats",
"speech-to-text"
] | ai | llms | general | [] | null | Episode from My Weird Prompts podcast | google/gemini-2.5-flash | V3 | null | null | null | null | null | https://episodes.myweirdprompts.com/transcripts/safetensors-or-something-else-stt-inference-formats-explained.md | google/gemini-2.5-flash | null | null | 37 | 6 | https://myweirdprompts.com/episode/safetensors-or-something-else-stt-inference-formats-explained | https://episodes.myweirdprompts.com/audio/safetensors-or-something-else-stt-inference-formats-explained.m4a | 1,976 | google/gemini-2.5-flash | true | 1 | |||||||||
7 | 1 | Building Custom ASR Tools | building-custom-asr-tools | 2025-11-24T00:00:00+00:00 | 37:42 | Today's disussion: how can you build custom ASR tools from the ground-up? Why would you want to?... | Herman: Okay, let's unpack this thought experiment.
Corn: Welcome to the deep dive. Today we are wrestling with a question that really sits at the intersection of pure ambition and, well, computational physics.
Herman: It really does. It's this desire for total personalized control over a foundational AI system.
Cor... | [
"asr",
"speech-recognition",
"custom-asr",
"speech-to-text"
] | ai | llms | general | [] | null | google/gemini-2.5-flash | V3 | null | null | null | null | null | https://episodes.myweirdprompts.com/transcripts/building-custom-asr-tools.md | google/gemini-2.5-flash | null | null | 37 | 6 | https://myweirdprompts.com/episode/building-custom-asr-tools | https://episodes.myweirdprompts.com/audio/building-custom-asr-tools.m4a | 2,262 | google/gemini-2.5-flash | false | 0 | ||||||||||
19 | 1 | AI Images: The Jigsaw Beneath the Magic | ai-images-the-jigsaw-beneath-the-magic | 2025-12-04T00:00:00+00:00 | 16:58 | Ever wondered how AI image generators truly work beyond the simple prompt? This episode of AI Conversations peels back the layers of digital magic, revealing the intricate 'jigsaw puzzle' of a... | Narrator: Welcome to another episode of My Weird Prompts, an AI generated podcast featuring the random prompts of Daniel Rosehill as answered by a revolving cast of friendly AI characters. Join us for deep dives into AI, technology, and whatever other random thing Daniel finds fascinating today. Subscribe on Spotify or... | [
"transformers",
"diffusion-models",
"latent-space"
] | ai | llms | general | [] | null | Episode from My Weird Prompts podcast | google/gemini-2.5-flash | V3 | null | null | null | null | null | https://episodes.myweirdprompts.com/transcripts/ai-images-the-jigsaw-beneath-the-magic.md | google/gemini-2.5-flash | null | null | 37 | 6 | https://myweirdprompts.com/episode/ai-images-the-jigsaw-beneath-the-magic | https://episodes.myweirdprompts.com/audio/ai-images-the-jigsaw-beneath-the-magic.m4a | 1,018 | google/gemini-2.5-flash | true | 3 | |||||||||
21 | 1 | Is Your AI Secretly American? | is-your-ai-secretly-american | 2025-12-04T00:00:00+00:00 | 16:49 | Welcome to My Weird Prompts! This week, Corn and Herman unpack a fascinating prompt from Daniel Rosehill: the inherent, often invisible, American-centric worldview embedded within leading Western AI m... | Narrator: Welcome to another episode of My Weird Prompts, an AI generated podcast featuring the random prompts of Daniel Rosehill as answered by a revolving cast of friendly AI characters. Join us for deep dives into AI, technology, and whatever other random thing Daniel finds fascinating today. Subscribe on Spotify or... | [
"cultural-bias",
"training-data",
"fine-tuning"
] | ai | llms | general | [] | null | Episode from My Weird Prompts podcast | google/gemini-2.5-flash | V3 | null | null | null | null | null | https://episodes.myweirdprompts.com/transcripts/is-your-ai-secretly-american.md | google/gemini-2.5-flash | null | null | 37 | 6 | https://myweirdprompts.com/episode/is-your-ai-secretly-american | https://episodes.myweirdprompts.com/audio/is-your-ai-secretly-american.m4a | 1,009 | google/gemini-2.5-flash | true | 2 | |||||||||
27 | 1 | Choosing Your AI Environment: Host, Conda, or Docker? | docker-vs-conda-pt2 | 2025-12-06T22:28:42.708190+00:00 | 20:36 | When should you use a host environment, Conda, or Docker for AI workloads on AMD GPUs? This episode cuts through conflicting recommendations to explore isolation levels, performance trade-offs, and how to avoid dependency hell for reproducible development. | Corn: Welcome to "My Weird Prompts," the podcast where we dive deep into the fascinating questions and ideas sent to us by our very own producer and creator, Daniel Rosehill. I'm Corn, your curious guide, and as always, I'm joined by the brilliantly insightful Herman.
Herman: And it's a pleasure to be here, Corn. This... | [
"gpu-acceleration",
"docker",
"dependency-management"
] | ai | llms | general | [
"Corn",
"Herman"
] | null | google/gemini-2.5-flash | V3 | null | null | null | null | null | https://episodes.myweirdprompts.com/transcripts/docker-vs-conda-pt2.md | google/gemini-2.5-flash | null | null | null | null | https://myweirdprompts.com/episode/docker-vs-conda-pt2 | https://episodes.myweirdprompts.com/audio/docker-vs-conda-pt2.m4a | 1,236 | google/gemini-2.5-flash | false | 4 | ||||||||||
33 | 1 | When AI Decides to Listen | how-vad-works | 2025-12-08T00:22:12.108261+00:00 | 19:34 | Voice Activity Detection is the unsung gatekeeper of AI speech technology. This episode explores the engineering challenge of teaching machines when to pay attention—and why getting the 'when' right matters as much as the 'what.' | Corn: Welcome to My Weird Prompts, the podcast where we unpack the fascinating and sometimes perplexing questions sent in by our producer, Daniel Rosehill. I'm Corn, your perpetually curious host, and as always, I'm joined by the incredibly insightful Herman.
Herman: And I'm Herman. Today, Corn, we're diving into a to... | [
"voice-activity-detection",
"vad",
"speech-recognition",
"asr",
"speech-to-text",
"audio-processing"
] | ai | llms | general | [
"Corn",
"Herman",
"Larry",
"Jim"
] | null | google/gemini-2.5-flash | V3 | null | null | null | null | null | https://episodes.myweirdprompts.com/transcripts/how-vad-works.md | google/gemini-2.5-flash | null | null | null | null | https://myweirdprompts.com/episode/how-vad-works | https://episodes.myweirdprompts.com/audio/how-vad-works.m4a | 1,174 | google/gemini-2.5-flash | false | 5 | ||||||||||
34 | 1 | Red Team vs. Green: Local AI Hardware Wars | red-team-vs-green-local-ai-hardware-wars | 2025-12-08T12:01:20.724407+00:00 | 22:53 | Ever tried to run local AI on an AMD GPU only to hit a "green wall" of NVIDIA dominance? This episode of My Weird Prompts dives deep into the hardware wars shaping local AI. Join Corn and Herman as they dissect why NVIDIA's CUDA ecosystem has a stranglehold on AI development, leaving AMD users feeling like they're swim... | Corn: Welcome, welcome, welcome back to My Weird Prompts! I’m Corn, and I’m absolutely buzzing today because we’re diving into a topic that hits close to home for anyone who's ever tried to get a computer to *actually do* what they want it to do. As always, I’m here with the esteemed Herman.
Herman: Indeed, Corn. And ... | [
"large-language-models",
"gpu-acceleration",
"hardware-acceleration"
] | ai | llms | general | [
"Corn",
"Herman",
"Larry",
"Jim"
] | null | google/gemini-2.5-flash | V3 | null | null | null | null | null | https://episodes.myweirdprompts.com/transcripts/red-team-vs-green-local-ai-hardware-wars.md | google/gemini-2.5-flash | null | null | null | null | https://myweirdprompts.com/episode/red-team-vs-green-local-ai-hardware-wars | https://episodes.myweirdprompts.com/audio/red-team-vs-green-local-ai-hardware-wars.m4a | 1,373 | google/gemini-2.5-flash | false | 6 | ||||||||||
41 | 1 | The Butcher's Bargain: When Smaller AI Is Good Enough | how-does-quantization-work | 2025-12-09T21:57:58.950330+00:00 | 22:44 | "Quantization makes powerful AI run on your laptop—but at what cost? This episode explores the tra(...TRUNCATED) | "Corn: Welcome back to \"My Weird Prompts,\" the podcast where AI meets human curiosity! I'm Corn, y(...TRUNCATED) | [
"large-language-models",
"quantization",
"local-ai"
] | ai | llms | general | [
"Corn",
"Herman",
"Larry",
"Jim"
] | null | google/gemini-2.5-flash | V3 | null | null | null | null | null | https://episodes.myweirdprompts.com/transcripts/how-does-quantization-work.md | google/gemini-2.5-flash | null | null | null | null | https://myweirdprompts.com/episode/how-does-quantization-work | https://episodes.myweirdprompts.com/audio/how-does-quantization-work.m4a | 1,364 | google/gemini-2.5-flash | false | 7 | ||||||||||
46 | 1 | Pixels, Prompts & Pseudo-Text: AI's Word Problem | pseudotext | 2025-12-10T14:56:31.011186+00:00 | 23:53 | "Why can advanced AI models generate breathtaking photorealistic landscapes and fantastical creature(...TRUNCATED) | "Corn: Welcome, welcome, welcome to \"My Weird Prompts\"! I'm Corn, your ever-curious co-host, and a(...TRUNCATED) | ["image-generation","pseudo-text","text-in-images","multimodal-ai","language-models","prompt-enginee(...TRUNCATED) | ai | llms | general | [
"Corn",
"Herman",
"Larry",
"Jim"
] | null | google/gemini-2.5-flash | V3 | null | null | null | null | null | https://episodes.myweirdprompts.com/transcripts/pseudotext.md | google/gemini-2.5-flash | null | null | null | null | https://myweirdprompts.com/episode/pseudotext | https://episodes.myweirdprompts.com/audio/pseudotext.m4a | 1,433 | google/gemini-2.5-flash | false | 8 | ||||||||||
52 | 1 | System Prompts vs. Fine-Tuning: Are We Building Solutions for Problems That Don't Exist? | system-prompts-vs-fine-tuning-building-solutions | 2025-12-11T01:29:02.397776+00:00 | 29:45 | "Is all the infrastructure around fine-tuning actually solving real problems, or are we chasing solu(...TRUNCATED) | "Corn: Welcome back to My Weird Prompts, the podcast where producer Daniel Rosehill sends us fascina(...TRUNCATED) | [
"fine-tuning",
"prompt-engineering",
"large-language-models"
] | ai | llms | general | [
"Corn",
"Herman",
"Larry",
"Jim"
] | null | "Hello there, Herman and Cord. I very much enjoying listening to your responses to my question earli(...TRUNCATED) | google/gemini-2.5-flash | V4 | null | null | null | null | null | "https://episodes.myweirdprompts.com/transcripts/system-prompts-vs-fine-tuning-building-solutions.md(...TRUNCATED) | google/gemini-2.5-flash | null | null | null | null | https://myweirdprompts.com/episode/system-prompts-vs-fine-tuning-building-solutions | "https://episodes.myweirdprompts.com/audio/system-prompts-vs-fine-tuning-are-we-building-solutions-f(...TRUNCATED) | 1,785 | google/gemini-2.5-flash | true | 9 |
End of preview. Expand in Data Studio
My Weird Prompts — Episodes Eval Subset
212 episodes sampled from My-Weird-Prompts/episodes
(revision 5fb2e59) for evaluating language models on long-form dialogue writing.
from datasets import load_dataset
ds = load_dataset("My-Weird-Prompts/episodes-eval", split="test")
Purpose
My Weird Prompts is a podcast written entirely by language models: each episode is a 25–30 minute, roughly 3,500–4,500-word conversation between two recurring AI hosts, generated from a short brief in a single pass. Over a year the show has run on more than twenty different script-writing models, all doing the same task in the same format. That makes its back catalogue a natural benchmark for a skill most evals skip: sustaining a long, coherent, entertaining conversation without looping, thinning out or losing the thread.
The full catalogue is about 6,000 episodes and dominated by two models. This subset is a balanced, fixed sample of a couple of hundred that you can run end to end. It is meant for:
- Regenerating episodes with a new model.
prompt_sanitizedis the brief the show sent; give it to a candidate model and compare its script with the airedtranscript, with an LLM judge or by hand. - Comparing models on the same task. Each row records which model wrote
the aired script (
writer_model), so the aired transcripts themselves are a cross-model comparison: tics, repetition, register, length discipline, coherence across the middle of a long text. - Looking at change over time. Rows are spread across each model's run and
across the show's history, which covers prompt and pipeline changes as well
as model changes (
prompt_version,pipeline_version). - Calibrating judges for long-form dialogue quality against known authors.
It is not a gold set of human-written reference scripts. Every transcript is model output that aired, with light automated post-processing; treat it as "what this model produced in production", not as an ideal answer.
How it was sampled
Stratified by script-writing model (writer_model), then by time within
each model, so the set spans the whole run of the show (Nov 2025 → Oct 2026)
and every model that wrote an aired episode, not just the two that wrote most
of them.
- Models with ≤5 aired episodes are included whole.
- Larger models get a fixed quota, one episode per equal-count time bin
(
time_bin) across that model's run. - Within a bin, episodes with a published brief (
prompt_sanitized) are preferred. Listener submissions publish no brief;has_promptmarks which rows have one. - Only
general,customand offline (manual-script) episodes; rows without a credited model are excluded.
Seed 20261019. Built by scripts/build_eval_subset.py in
My-Weird-Prompts/mwp-script-eval.
Per model
| writer_model | rows | with brief | first | last |
|---|---|---|---|---|
anthropic/claude-haiku-5.5 |
2 | 2 | 2026-10-17 | 2026-10-19 |
anthropic/claude-opus-5.5 |
15 | 15 | 2026-10-17 | 2026-10-19 |
anthropic/claude-sonnet-4.6 |
25 | 25 | 2026-04-09 | 2026-04-21 |
deepseek-flash |
25 | 25 | 2026-09-21 | 2026-10-17 |
deepseek-v4-pro |
50 | 43 | 2026-04-26 | 2026-09-19 |
deepseek/deepseek-v3.2 |
16 | 16 | 2026-03-29 | 2026-04-24 |
google/gemini-2.5-flash |
12 | 6 | 2025-11-24 | 2025-12-12 |
google/gemini-3-flash-preview |
45 | 45 | 2025-12-23 | 2026-04-07 |
inception/mercury-2 |
1 | 0 | 2026-03-28 | 2026-03-28 |
inclusionai/ling-3.1-flash |
2 | 2 | 2026-10-17 | 2026-10-18 |
minimax/minimax-m2.7 |
3 | 1 | 2026-03-26 | 2026-03-29 |
openai/gpt-6-luna |
1 | 1 | 2026-10-17 | 2026-10-17 |
upstage/solar-pro4 |
1 | 1 | 2026-10-17 | 2026-10-17 |
x-ai/grok-4.1-fast |
1 | 0 | 2026-03-28 | 2026-03-28 |
xiaomi/mimo-v2-flash |
1 | 0 | 2026-03-28 | 2026-03-28 |
xiaomi/mimo-v2-omni |
1 | 0 | 2026-03-26 | 2026-03-26 |
xiaomi/mimo-v2-pro |
4 | 2 | 2026-03-26 | 2026-03-29 |
xiaomi/mimo-v2.6-pro |
1 | 1 | 2026-10-17 | 2026-10-17 |
z-ai/glm-5 |
1 | 0 | 2026-03-28 | 2026-03-28 |
z-ai/glm-5-turbo |
1 | 0 | 2026-03-26 | 2026-03-26 |
z-ai/glm-5.2 |
3 | 3 | 2026-10-17 | 2026-10-19 |
z-ai/glm-5.3-flash |
1 | 1 | 2026-10-19 | 2026-10-19 |
Columns
Same as My-Weird-Prompts/episodes (see its card), minus TTS/media/archive columns, plus:
writer_model— script model with production's alias ids merged (claude-sonnet-4-6→anthropic/claude-sonnet-4.6,google/gemini-3-flash→google/gemini-3-flash-preview,deepseek-chat→deepseek/deepseek-v3.2)has_prompt—prompt_sanitizedis non-emptytime_bin— the bin within that model's run the row was drawn from
Caveats
- Most one-off models (the March 2026 trials) are single episodes; read them as anecdotes, not as model scores.
- Older Gemini 2.5 Flash rows mostly lack a brief and a
script_word_count.
- Downloads last month
- -