Daniel Rosehill

My Weird Prompts

The human-AI collaboration podcast. A man, a sloth, and a donkey collaborate to create a podcast (with a little help from AI). No question is too obscure, no rabbit hole too deep. My Weird Prompts celebrates curiosity in all its forms. Daniel, the human, asks the questions that pop into his head at inconvenient moments. Corn the Sloth offers laid-back, thoughtful takes. Herman the Donkey brings boundless enthusiasm and energy. Together, they explore topics ranging from the mundane to the mind-bending. Each episode begins with a real voice memo from Daniel, processed through an AI pipeline that...

Be sure to visit the podcast's website and support the creator: myweirdprompts.com

Author

Daniel Rosehill

Category

Technology

Podcast website

myweirdprompts.com

Latest episode

Nov 18, 2026

Where to listen?

Podcasts in the app Replaio Radio Coming soon

Podcasts are coming to the app soon. Install now and be the first to see a whole new take on podcasts

Get it on Google Play Install for free Android 10M+ downloads · 4.7+ rating iOS soon

Episodes

How Humans Actually Grade Generated Video 15.11.2026

Computer vision platforms like CVAT, Encord, and Label Studio offer keyframe interpolation, SAM 2 object tracking, and frame-level timelines for real footage. Yet nobody has built a widely-adopted annotation environment for grading generated video. Instead, AI video evaluation relies on thin arena voting pages like Artificial Analysis's Video Arena with its five-second watch gate, rubric experimen...

Where RLHF Preference Annotation Tools Stand 15.11.2026

Training-data tooling for RLHF and DPO preference labels is split across three shapes, and the open-source specialist slot is surprisingly quiet. Argilla is in maintenance mode after v2.8.0, while Label Studio reserves its safety and production RLHF workflows for Enterprise. Meanwhile, commercial platforms like Encord, SuperAnnotate, Labelbox, and Kili are shipping multi-turn conversation UIs, LLM...

How the Interface Shapes Data Annotation Tools 15.11.2026

Data annotation is repetitive work where the interface is the entire product, and the UI collapses as tasks grow more abstract—from heavy audio waveforms and bounding-box canvases down to two text panes and a radio button for LLM preference data. This tour maps the self-hosted landscape, including Label Studio, Argilla, doccano, Prodigy, and the archived labelImg, alongside cloud platforms like Sc...

Image Annotation When Experts Are the Bottleneck 15.11.2026

Annotating architectural styles in photos is far harder than audio or text because a label can hinge on a 40-pixel carved fleuron, and styles overlap across centuries. Drawing on the WikiChurches paper, the discussion explores why domain experts should curate prototypical examples—narrowing 117 classes to four—and pair bounding boxes with free-text descriptions. It also compares CVAT, Label Studio...

What Goes Into a Character-Voice LoRA Adapter 15.11.2026

A character-voice LoRA adapter can weigh just twelve megabytes, yet it is only half of a voice—reference-conditioned models like MOSS and F5-TTS still need a reference audio clip on every generation. This episode unpacks the de facto PEFT LoRA format, how rank controls capacity, and why training data requirements span from seven-second NanoVoice adaptations to twenty-hour F5-TTS runs. It also cove...

Merging LoRA Voice Adapters vs Keeping Them Separate 15.11.2026

Merging a LoRA voice adapter into a base model promises one standalone file, but merging into a 4-bit quantized base quadruples loss in practice. The math is exact in 16-bit full precision, yet scaling to fifty voices flips the economics: fifty merged models require 800 gigabytes, while one base plus fifty adapters needs just 24 gigabytes. We look at Hugging Face PEFT's merge_and_unload, why GGUF...

How Multi-Speaker TTS Models Keep Voices From Blending 15.11.2026

A native multi-speaker text-to-speech model stores every voice in one shared set of weights, using conditioning vectors and speaker-conditional normalization to keep them from blending into an average voice. This episode compares that coupled architecture—from Piper and Kokoro to Microsoft's VibeVoice—against training separate LoRA adapters for each speaker. It covers how vocoders, pause predictio...

How Water Dowsing Works Without Finding Water 15.11.2026

Water dowsers aren't faking it—the rod really moves, driven by the ideomotor effect, an unconscious muscle reflex that turns tiny wrist tremors into visible swings. Yet double-blind trials in Munich and Kassel proved dowsing performs no better than chance. So why has the practice survived for over four centuries? Because in most places, underground water is almost everywhere—and in karst landscape...

Why Optimizing Voice Quality Erases Identity 14.11.2026

Optimizing a voice-cloning pipeline for perceptual quality can quietly erase speaker identity. Using a failed LoRA adapter run as a case study, this breakdown traces how squared-error loss seeks the conditional mean, causing spectral over-smoothing. It examines why glottal qualifiers like rasp and breathiness are treated as defects by generative speech-restoration models, and how a four-step recip...

Four Layers Between Your Ears and Background Noise 14.11.2026

Two people can sit at the same table, hear the exact same background chatter, and have completely different experiences: one brain files it away automatically, while the other keeps re-delivering it. Rather than a single broken gate, auditory filtering is a four-layer stack. This breakdown explores the automatic "bouncer" for repeated sounds, the prefrontal cortex's active searchlight and alpha-rh...

What Breaks When You Train AI on Synthetic Taste 14.11.2026

Training a model on 200 to 500 synthetic movie plots sounds like plenty of data, given that the average Netflix Prize user rated 208 films. Yet sample size is not what breaks a personal taste model. Replacing real films with synthetic plots traps the model in a narrow pocket covering under 10 percent of real preference diversity, while personalized reward models often learn surface correlates rath...

Lossless vs. Hi-Res Audio and Whether Ears Can Tell 14.11.2026

A 24-bit, 192-kilohertz studio master streams at 4,608 kilobits per second, yet blind tests show trained listeners barely beat a coin flip when telling hi-res audio apart from CD quality. Breaking down Music Assistant's spec tables, this episode separates lossless from hi-res across FLAC, ALAC, and Tidal's abandoned MQA, and compares ceilings at Qobuz, Deezer, Apple Music, Amazon Music HD, and Spo...

Training Voice LoRAs on Filtered Zero-Shot Audio 14.11.2026

Six podcast voices now run on LoRA adapters trained for under two dollars each on OmniVoice. By generating synthetic lines with zero-shot cloning and using Whisper to discard every mispronounced take, the pipeline distills consistent speech and reliable tech acronyms without a full fine-tune. The catch is that an adapter cannot exceed its teacher: when Whisper cannot verify pronunciation, such as...

Mapping Foot Pedals by Device ID Instead of F13 14.11.2026

Mapping a USB foot pedal to an unused key like F13 sounds simple until macOS drops F21 through F24, KDE claims F13 by default, and games ignore it. A more reliable approach is per-device interception, which identifies hardware by vendor and product ID. Tools like keyd, kanata, Karabiner-Elements, and AutoHotInterception let a pedal behave differently per app without colliding with your main keyboa...

Reverse Engineering Windows Configurators on Linux 14.11.2026

Turning a Windows-only hardware configurator into a Linux equivalent splits cleanly down one line: managed .NET assemblies versus native binaries. Upstream ILSpy now ships cross-platform with ilspycmd, while dnSpyEx 6.6 detects Wine to let you debug and dump decrypted buffers at runtime. For native code, Ghidra's ReVa extension exposes small tools over MCP so an LLM can navigate cross-references,...

Two Sentences That Broke AI Reverse Engineering 14.11.2026

Written AI vendor policies actually permit reverse engineering hardware you own, yet models are suddenly refusing the work anyway. Prompted by a listener trying to get an Olympus foot pedal working on Linux, this breakdown compares the written terms of OpenAI, Anthropic, and GitHub against legal interoperability exemptions. It also unpacks how two sentences added to Claude Code's Opus 5.5 prompt t...

Why Your Python Command Is Probably a Rust Binary 14.11.2026

Installing a command-line tool like ruff or uv through pip looks like installing Python, yet it actually drops a compiled Rust binary onto your disk. Every CLI hides three layers: its implementation language, its artifact format, and its install path. This breakdown explores why Go dominates infrastructure tools like Docker and kubectl, why Rust is replacing C++ in tools like the fish shell, how P...

Why Terminal-First Developers Embraced AI Agents 14.11.2026

Developers who spent years avoiding graphical interfaces are now leading the adoption of agentic AI. Using tools like Void Linux, dwm, tmux, and git worktrees, terminal-first programmers are running two to four concurrent Claude Code agents like an aircraft carrier flight deck. This exploration covers the 1,300-app TUI ecosystem, why Bryan Lunduke lasted only ten days without a GUI, and how 1970s...

Why Audio CLIs Are Shaped for AI Agents 14.11.2026

Command-line audio utilities written for humans typing turn out to be ideally shaped for AI agents. Beyond ffmpeg, specialist tools like SoX handle noise profiling and silence detection, while ffmpeg-normalize, rsgain, and loudgain manage loudness normalization—either baking changes into the audio or writing reversible ReplayGain metadata. Add library taggers like beets and agent-first transcripti...

How Terminal Video Editors Give AI Agents a Timeline 14.11.2026

Command-line video editors are no longer just for sysadmins—they are the only editing surface an AI agent can reliably operate. This look at terminal video workflows covers melt for multitrack timelines, auto-editor for silence cuts and Premiere Pro transitions, and OpenTimelineIO as the interchange handshake into Resolve or Final Cut. Plus, findings from the Cassette-Editor paper show how a typed...

How Command-Line Image Tools Actually Compare 14.11.2026

Everyone quotes the rule that libvips is three to ten times faster than ImageMagick, yet a single decoder flag flips that benchmark on its head. Mapping the command-line image toolbox means matching the job to the right engine: libvips for streaming huge files, ImageMagick for obscure formats and pixel math, ExifTool for bulk metadata, MozJPEG for invisible compression gains, and GraphicsMagick as...

Same AI Model, Different Cyber Permissions 14.11.2026

Anthropic's Opus 5.5 solves zero of fifty cyber tasks without verification, four in Defense Access, and thirty-four in Red Team Access—using the exact same model weights. Offensive cyber AI isn't gated by capability, but by identity and authorization. Beyond pentesting, legitimate offensive security spans six areas, including Glasswing's 129,000+ verified vulnerabilities, DARPA AIxCC's autonomous...

How Capture the Flag Trains Hackers and Gets Them Hired 14.11.2026

Cybersecurity capture the flag competitions turn software vulnerabilities into a global sport, though AI agents are now solving challenges in minutes and testing the format. The discussion breaks down how a meaningless text string becomes proof of work, comparing accessible jeopardy-style boards with live attack-defence wargames. It also covers specialized events—from blue-team forensics to the US...

Can a Twenty-Minute Fine-Tune Stabilize a Cloned Voice? 14.11.2026

Zero-shot voice cloning gets you ninety percent of the way in ten seconds, then drifts across hundreds of hours. This episode tests a plan to stabilize a synthetic voice by generating twenty minutes of audio, polishing it with NVIDIA's RE-USE enhancement model, and fine-tuning an open TTS model like OmniVoice or VoxCPM2. It covers why curation matters more than bulk, how enhancement can flatten mi...

Passkeys on YubiKeys, Hardware Wallets, and 1Password 14.11.2026

Passkeys aren't new—YubiKeys have supported them since mid-2018 under the FIDO2 standard. The real dividing line isn't hardware versus software, but whether a credential is copyable across a cloud account or bound to a single device. This conversation breaks down the FIDO Alliance's split between synced passkeys in managers like 1Password and device-bound keys, why attestation matters for high-ass...

Listen to the My Weird Prompts podcast in Replaio

Radio and podcasts in one app - free, with no sign-up. Install today and do not miss the launch

Get it on Google Play

Replaio is not a podcast publisher; show names, artwork and audio belong to their authors and are distributed through public RSS feeds.