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
Podcast website
Latest episode
Nov 18, 2026
Where to listen?
Podcasts in the app Replaio Radio Coming soonPodcasts are coming to the app soon. Install now and be the first to see a whole new take on podcasts
Episodes
Why Even Solo Developers Face a GPU Shortage 09.11.2026 22:30
On-demand GPU rental capacity has quietly dried up, catching solo developers fine-tuning on a single L4 just as severely as hyperscalers. An analysis of surging memory costs, multi-year contracts, and discontinued older architectures reveals why providers are deprioritising on-demand users. In response, builders turn to public compute initiatives like NAIRR and EuroHPC, where free academic allocat...
Why Dual-Socket Servers Win on Dense AI Models 09.11.2026 26:03
Adding a second CPU socket offers extra memory channels and PCIe lanes, but non-uniform memory access dramatically shapes real-world performance. A detailed look at Intel Xeon and AMD EPYC platforms reveals why dense neural networks scale up while fine-grained mixture-of-experts architectures stall on cross-socket interconnects. Also covered: BIOS nodes-per-socket settings, the economics of decomm...
Build Servers Never Actually Build Anything 09.11.2026 22:15
Continuous integration terminology is full of misnomers, starting with the fact that build servers rarely compile code. Evaluating Woodpecker CI, Forgejo Runner, and self-hosted GitHub Actions runners reveals how build jobs actually move from git push to execution. Job routing relies on lease-based polling queues, label matching, and outbound-only network connections, while isolation models vary w...
Why Voiceprints Are Not an Interoperable File Format 09.11.2026 23:28
Voice biometrics often promises portable voiceprints, yet no vendor-neutral standard exists to exchange speaker models. The official ISO/IEC 19794-13 biometrics standard deliberately covers only raw audio packaging, leaving extraction and speaker embeddings outside its scope due to vendor trade secrets. Looking at efforts like Brno University of Technology's Voice Biometry Standardization Initiati...
Why Immigrants Speak Hebrish 09.11.2026 22:51
English-speaking immigrants in Israel speak an everyday dialect packed with untranslated Hebrew terms. While popular culture fixates on poetic idioms like firgun, real-world linguistic borrowing is driven by administrative necessity, preserving words like arnona and tofes because translating them loses vital structural information. From speech recognition benchmarks to cultural scripts, the vocabu...
Voice Isolation Before Speech Recognition 09.11.2026 20:45
Voice isolation models strip out competing speakers before speech reaches voice activity detection and speech-to-text engines. While models like VoxLock, Scicom, Hush, and ai-coustics' Quail Voice Focus dramatically improve transcriptions in noisy multi-speaker settings, running them on clean audio consistently degrades transcription accuracy. From automated enrollment heuristics and decoy recordi...
How a Knesset Government Can Form with Under 61 Votes 09.11.2026 27:13
Coalition formation after the election rests on parliamentary mathematics rather than pure majorities. An incoming Israeli government does not actually require sixty-one affirmative votes to pass its investiture vote, creating an opening for a fragile minority government sustained by external abstentions. Breaking down the certified Knesset seat distributions, this analysis explores how competing...
Why Speech Transducers Omit Common Symbols 09.11.2026 25:15
When speech models fail to produce common characters like email at-signs, the limitation stems from deep architectural tradeoffs rather than missing training data. RNN-T and TDT architectures calculate a four-dimensional joint tensor that scales directly with vocabulary size, making large vocabularies memory-prohibitive. This architectural constraint separates models into two camps: massive LLM-st...
Why Broadcast Headsets Fit Commentators but Not Voiceover 08.11.2026 21:20
Closed-cup broadcast headsets dominate sports commentary press boxes by locking mic distance and blocking crowd noise, but using one for home voiceover brings distinct trade-offs. A tour of legacy models from Sennheiser, Audio-Technica, Shure, and Beyerdynamic reveals why the category remains strictly analog XLR rather than USB, how proximity effect and clamping force challenge long narration, and...
Why Generative Audio Restoration Hallucinates Highs 08.11.2026 27:43
Modern audio restoration tools like NVIDIA's RE-USE and Tokyo's Sidon ditch traditional filtering to synthesize clean speech representations from scratch. While they effortlessly erase background baby cries and air conditioner hums, their underlying mechanics reveal a hidden catch. Recordings from cheaper phone mics often yield better synthesis datasets than filtered headsets because hardware-limi...
Why Blanket Forts Beat Expensive Portable Vocal Booths 08.11.2026 22:53
Portable vocal setups promise studio sound anywhere, but most home recording purchases disappoint because buyers confuse acoustic treatment with isolation. Blankets, clothes-filled closets, and commercial travel booths alter reflections inside a room without blocking outside traffic or barking dogs. The physics of heavy moving blankets outshine cheap acoustic panels, while parked cars offer surpri...
Why Voiceover Studios Obsess Over Repeatability 08.11.2026 24:52
Professional voiceover setups are engineered around physical consistency, ensuring recordings made months apart match seamlessly. This deep dive covers microphone positioning geometry, off-axis plosive management, eye-level script placement to eliminate head drift, and why tablet reflections matter. It also details the acoustics of sitting versus standing, hands-free punch-and-roll workflows in Re...
Why Tokenizer Vocabulary Size Is a Hardware Choice 08.11.2026 25:22
Before an AI model reads a word, a pre-trained tokenizer chops the text into pieces using byte pair encoding. While vocabulary size is often treated as an arbitrary convention, its optimal dimension varies dramatically depending on whether a model runs on-device or across datacenter GPUs. Looking into merge counts, multilingual additions from Llama 3 to Gemini, and recent findings on Zipf's law, t...
What a Cross-Entropy Loss of 1.8 Actually Means 08.11.2026 25:22
Training runs live and die by cross-entropy loss curves, yet few can translate a single loss number into plain terms. This breakdown walks through how a loss of 1.8 nats translates directly into an average perplexity of six branching choices, why language modeling mathematically equates to compression, and how differing tokenizers produce deceptive per-token numbers. It also explores why tools lik...
Splitting Frontier AI Across Thousands of GPUs 08.11.2026 30:44
Training frontier AI models is fundamentally a distributed plumbing challenge across thousands of accelerators. When weights, optimizer states, and activations vastly exceed single-chip memory, engineers must split workloads across data, tensor, pipeline, and expert parallelism. Balancing these techniques reveals hard trade-offs between network latency and memory bandwidth, showing why out-of-the-...
Why LLM Leaderboards Need Psychometrics 08.11.2026 27:41
Standard AI benchmarks treat every question as an equal test of intelligence, but educational testing abandoned raw score averages decades ago. Applying item response theory exposes how popular benchmarks like MMLU and HellaSwag have grown far too easy to separate top models. Adaptive testing methods such as ATLAS can evaluate models in dozens of questions instead of thousands while adjusting rank...
Why Simple Rules Beat Value Models in Reasoning RL 08.11.2026 31:47
Reinforcement learning with verifiable rewards powers modern reasoning models without requiring complex neural critics. By replacing a policy-sized value model with group-relative advantage estimation, algorithms like GRPO cut memory overhead and avoid reward hacking. A look at the inner mechanics of rule-based scoring, how crude outcome verification generates tens of thousands of reasoning tokens...
Why You Must Read AI Traces Before Writing Evals 08.11.2026 25:54
Standard advice suggests picking automated benchmarks and metrics first, but practitioners argue that effective AI evaluation requires reading raw interaction traces by hand. Discover how grounded theory methods like open and axial coding expose criteria drift and application-specific flaws before tools are deployed. The discussion covers practical trace sample sizes, how to spot upstream root cau...
Why Speech Recognition Confidence Scores Lie 08.11.2026 26:36
Speech recognition models like Whisper and Parakeet compute internal probabilities for every token, but turning those numbers into trustworthy word-level confidence is surprisingly brittle. Softmax saturation, pooling choices across sub-words, decoding fallbacks, and blank frames in CTC architectures distort confidence scores across the board. The mechanics behind word confidence calculations, why...
How Speech Models See Audio as an Image 08.11.2026 30:01
Speech recognition models like Whisper never actually hear audio as human listeners do. Before a neural network processes anything, incoming waveforms are resampled, cut into twenty-five-millisecond windows, and compressed into log-mel spectrograms. This breakdown walks through every stage of that front-end pipeline, exploring the 1937 five-person study behind the mel scale, the deliberate discard...
Teaching Speech Models to Choose Their Own Audio 08.11.2026 27:40
Human transcription takes over eight hours of manual work for every hour of recorded audio. Active learning promises to cut that burden by letting speech models select which unlabelled recordings are most informative to transcribe next. Yet simple uncertainty measures often select garbage noise rather than useful speech, and sophisticated loss predictions frequently struggle against random baselin...
Where Speech Models Actually Learn 08.11.2026 28:05
Speech model fine-tuning recipes usually decide which layers get gradients before training even begins. Research reveals that encoders already capture nuanced semantics, speaking styles, and accent features that decoders systematically discard. Examining empirical benchmarks across Whisper and Parakeet, this discussion unpacks why freezing the acoustic encoder degrades multilingual accuracy, how h...
Why Speech Models Forget Other Voices When Tuned on Yours 08.11.2026 21:40
Fine-tuning a speech recognition model on a single speaker's voice often quietly destroys its ability to transcribe other accents, longer audio, and basic formatting. Examining evidence from Whisper adaptations in Hebrew and Swiss German, this audit covers how catastrophic forgetting erodes timestamps and structure even when target accuracy improves. It breaks down why general regression testing s...
What Audio Models Learn Before Fine-Tuning 08.11.2026 27:02
Speech encoders arrive at fine-tuning with tens of thousands of hours of audio structure already baked into their weights. This examination of internal representations contrasts self-supervised models like wav2vec 2.0, HuBERT, and w2v-BERT against weakly supervised systems like Whisper. It explores layer-by-layer probing of tongue position and word boundaries, the resurgence of speaker identity in...
Agent Skills Break When Secrets Leave Your Laptop 08.11.2026 22:43
Agent skills pack instructions and code into portable folders, but credentials break the moment workloads migrate away from local machines. While tools like the GitHub CLI prove that agents can drive workflows without dedicated servers, secrets lack a native home in the Agent Skills specification. Examining divergent credential approaches across Claude Code cloud sessions, Codex cloud, and GitHub...
Similar podcasts
Replaio is not a podcast publisher; show names, artwork and audio belong to their authors and are distributed through public RSS feeds.