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...
Vizitează neapărat site-ul podcastului și susține-i creatorul: myweirdprompts.com
Autor
Daniel Rosehill
Categorie
Site-ul podcastului
Cel mai nou episod
19 nov. 2026
Unde asculți?
Podcasturi în aplicație Replaio Radio În curândPodcasturile ajung în curând în aplicație. Instaleaz-o acum și fii primul care descoperă o abordare complet nouă a podcasturilor
Episoade
Voice Engineering and the TTS Recording Matrix 16.11.2026 27:04
Retraining OmniVoice LoRA adapters on older Chatterbox synthetic speech with a light EQ pass worked surprisingly well, raising a deeper question: what does deliberate voice engineering look like? This conversation explores how LoRA handles off-distribution audio, why corrective EQ on synthetic output differs from destructive EQ on human capture, and what an Interspeech Greek TTS study reveals abou...
Building a Toddler-Proof Linux Laptop With One App 16.11.2026 24:06
A listener wants to turn an old laptop into a safe, interactive machine for his sixteen-month-old son. The hardest technical piece already exists in Debian stable: BamBam, a baby-masher app that runs as a dedicated graphical session with no desktop underneath to break. Exploring the difference between input lockers like Hands Off and Catlock, the graveyard of kid-focused Linux distros like DoudouL...
Why a $30 Cardputer Can't Replace a Pocket Linux PC 16.11.2026 20:15
A $30 M5Stack Cardputer ADV looks like a tiny computer, but it runs firmware without an operating system, so it can't run a browser or act as a USB microphone. Mapping the world of cyberdecks and pocket computers reveals a sharp hardware divide between microcontroller handhelds—like the Cardputer ADV, LILYGO T-Deck Plus, and ClockworkPi PicoCalc—and full Linux machines such as the CardputerZero, u...
Using LLMs as Annotators for Silver Training Data 16.11.2026 23:38
When an LLM labels data for a downstream classifier, it isn't acting as a judge or classic weak supervision—it's an LLM-as-annotator producing silver labels. Exploring a Romanized Hebrew span-detection pipeline built with Claude Opus, this episode separates few-shot annotation and human-AI refinement loops from Snorkel-style labeling functions. It also covers the Alternative Annotator Test for spo...
How Canopy Searches AI Skills and Tools on Demand 16.11.2026 36:36
Nine hundred and twenty-seven tool names cost 20,010 tokens at the start of every session, even with schemas deferred. Canopy is a self-hosted AI gateway built to fix that by putting skills, plugins, tools, and standing notes behind a single searchable MCP endpoint. We look at why tool discovery exists in Anthropic's API while skill discovery stalls in Claude Code and MCP's SEP 2640, how Canopy's...
Same 100 Hours, 98% vs 6.74% Word Error in ASR 16.11.2026 28:49
Training a speech recognition model from scratch on 100 hours of LibriSpeech audio can produce a 98.4% word error rate in an attention encoder-decoder like SoralASR-23M-Preview, or 6.74% in k2-fsa's pruned stateless transducer. The difference comes down to monotonic alignment in transducers and CTC versus unguided attention. We also bracket the real compute costs of from-scratch runs like Zipforme...
Separating Voice From Pronunciation in a TTS Model 16.11.2026 18:58
Can a text-to-speech model keep a speaker's Hebrew pronunciation while dropping their Irish accent? Over three days building Meorav, an English-with-Hebrew model, synthetic phonemes failed on gutturals and Seed-VC voice conversion leaked the original accent's rhythm. A forty-clip x-vector pilot revealed that pronunciation lives in the model weights, separate from speaker identity. By splitting off...
How a Wooden Mouse Proved Machine Learning Worked 16.11.2026 21:33
Before machine learning had a name, Claude Shannon built a wooden mouse named Theseus that actually learned. Unlike earlier theoretical landmarks from McCulloch and Pitts in 1943 or Alan Turing in 1950, Theseus used telephone relays under a movable maze to remember paths—and forget them when walls changed—to help AT&T model phone switching. The story also covers Marvin Minsky's piano-sized SNARC,...
How MP3 Deletes 90% of a Song Without You Noticing 16.11.2026 27:57
MP3 shrinks CD audio from 1.41 megabits per second down by 75 to 95 percent, yet trained listeners often cannot tell the difference. Tracing the format from Dieter Seitzer's rejected 1970s telephone patent to Karlheinz Brandenburg's work on auditory masking, this story unpacks how psychoacoustics replaced raw waveform coding like ADPCM. It also covers early rivals like Sony's MiniDisc and Philips'...
How One Matrix Trick Powers Text, Image, and Voice LoRAs 16.11.2026 25:45
A single matrix equation—freezing a model's weights and learning a low-rank update as two tiny matrices—powers custom voices, image styles, and language models alike. The mechanism adds zero inference latency when merged, yet shrinks a 350-gigabyte GPT-3 checkpoint to 35 megabytes and a Stable Diffusion fine-tune to just over three. From rank-two bottlenecks and alpha scaling to why text, image, a...
Teaching Five AI Voices Israeli Hebrew 16.11.2026 28:53
A speech model can nail an English sentence and still butcher a single Hebrew word in the middle of it. Meorav Multi Voice solves that join by training Qwen3-TTS on 461 code-switched takes where only the Hebrew spans are graded, keeping the pronunciation without leaking the speaker's voice. Five synthetic accents from OmniVoice—American, British, and Australian—share that Hebrew training for about...
How Far Postgres Actually Scales in Production 16.11.2026 26:35
ChatGPT's backend serves 800 million users on a single unsharded Postgres primary, proving that database ceilings are rarely about raw capacity. Instead, production teams hit operational limits like vacuum, connection counts, and lock queues. This episode examines how OpenAI keeps one writer alive with nearly fifty read replicas and five-second schema timeouts, alongside why Instagram and Notion c...
How the Kaldi Team Pivoted to OmniVoice 16.11.2026 25:18
Speech recognition and speech synthesis have historically been separate worlds, yet the k2-fsa team at Xiaomi crossed over to build OmniVoice in just ten months. Tracing the lineage from Daniel Povey's Kaldi toolkit and Next-gen Kaldi's PyTorch stack to Han Zhu's ZipVoice, this deep dive unpacks how data pipelines proved direction-agnostic. It also breaks down OmniVoice's single-stage masked diffu...
When Speech Enhancement Erases Podcast Voice Character 16.11.2026 30:25
After Resemble AI retired its Chatterbox text-to-speech model in August, migrating the show's voices to OmniVoice seemed straightforward. But running the audio through NVIDIA's RE-USE speech enhancement model before training LoRA adapters backfired, stripping away the rough edges that gave the characters their personalities. The result was technically clean, completely forgettable, and a reminder...
When Merging LoRA Adapters Stops Being Exact 16.11.2026 23:40
Merging a LoRA adapter into its base model is exact in algebra, but finite-precision hardware tells a different story. Folding a half-mebibyte adapter into a 32-mebibyte matrix removes a serving branch rather than shrinking the model, and trades away swappability. The conversation unpacks how bf16 rounding, PEFT's hidden fp32 upcasting, and quantization introduce silent drift—as seen in a document...
How Label Studio's XML Configs and Storage Work 15.11.2026 26:48
Label Studio powers everything from bounding boxes to audio transcription with one design decision: an XML file that wires data types to annotation controls. This walkthrough covers how HumanSignal's open-source tool separates free individual labeling from paid team coordination, and what self-hosting to the Hugging Face Hub actually looks like. Expect the real architecture behind cloud storage sy...
Why CVAT Beats Generalists at Video Annotation 15.11.2026 21:55
CVAT has fewer GitHub stars than Label Studio, yet it remains the reference tool for video annotation. Originating inside Intel in 2017, its narrow focus on computer vision delivers keyframe interpolation, Outside-Merge-Split track primitives, and 3D point-cloud support that generalists lack. This deep dive examines how CVAT's architecture handles model-assisted labelling via Nuclio and CVAT AI Ag...
Small Labeling Tools That Outship Big Platforms 15.11.2026 22:27
Small, single-machine annotation tools are shipping faster than heavyweight platforms like Label Studio and CVAT. While LabelImg sits archived with 25,000 stars, desktop apps like labelme and X-AnyLabeling are adding SAM 3.1 video tracking, document parsing, and 3D point clouds in weeks. We explore six reasons solo developers and small teams skip big platforms—from local privacy and plain-text fil...
Quantization Quietly Deletes Fine-Tuned Skills 15.11.2026 25:17
A fine-tuned text-to-speech model called Meorav handled English-Hebrew code-switching at full precision, then lost the skill completely after quantization while still sounding normal. Exploring why compression quietly erases low-margin fine-tuned behaviors, this conversation unpacks outlier weights, signal-to-noise ratios, and why aggregate metrics like perplexity miss the damage. From Llama-3.1-8...
How Humans Actually Grade Generated Video 15.11.2026 26:07
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 24:48
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 20:29
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 21:56
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 29:31
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 21:54
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...
Podcasturi asemănătoare
Replaio nu este editorul podcasturilor; numele emisiunilor, coperțile și materialul audio aparțin autorilor lor și sunt distribuite prin fluxuri RSS publice