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...

Husk at besøge podcastens hjemmeside og støtte skaberen: myweirdprompts.com

Forfatter

Daniel Rosehill

Kategori

Technology

Podcastens hjemmeside

myweirdprompts.com

Seneste episode

19. nov. 2026

Hvor kan du lytte?

Podcasts i appen Replaio Radio Kommer snart

Podcasts kommer snart til appen. Installer nu, og vær den første til at se en helt ny tilgang til podcasts

Hent den på Google Play Installer gratis Android 10 mio.+ downloads · 4,7+ i bedømmelse iOS snart

Episoder

Splitting LLM Layers Across AMD and Nvidia GPUs 29.10.2026

Running an LLM too large for a single machine is possible by pooling memory across separate AMD and Nvidia computers. Tools like llama.cpp's RPC backend, HiveLink, and Parallax let mixed devices hold contiguous layer blocks and pass intermediate activations over a local network. This setup avoids vendor driver conflicts by dividing work rather than weights, but network latency, compute imbalance,...

Why Toddlers Say No Before They Can Explain Why 29.10.2026

Toddlers often master refusal long before they can articulate reasons, negotiate, or regulate disappointment. Examining early language development reveals that negation is relational rather than a label for objects, serving as a powerful communication shortcut. The notorious toddler refusal phase reflects uneven maturation across planning, inhibitory control, and speech. Cultural caregiving patter...

How vLLM Squeezes More Requests into GPU Memory 29.10.2026

Language model serving often stalls not from slow GPU arithmetic, but from greedy memory allocators reserving space for tokens that never arrive. Explore how UC Berkeley's vLLM project borrowed virtual memory and paging concepts from operating systems to create PagedAttention. By breaking KV caches into non-contiguous physical blocks, sharing prompt states through copy-on-write, and enabling prefi...

Training Text Models to Know When Not to Edit 29.10.2026

Training a text-cleanup model exclusively on flawed examples risks teaching it that every input requires intervention. Examining small models like Superwhisper's 600-million parameter S1-mini reveals the subtle challenge of training restraint. Balancing edited pairs with unchanged identity targets prevents compulsive rewriting while maintaining punctuation and formatting. Designing clean inputs, s...

Why Live Video Generation Still Struggles With Timing 29.10.2026

High frame rates do not guarantee instant video transformation. Live video diffusion models must abandon access to future frames, rely on causal attention, and compress generation into a few distilled steps. This breakdown examines the engineering behind systems like LiveEdit and StreamDiffusionV2, exploring why separating tracking from rendering matters, how error accumulation alters identities o...

Why Pitch Shifting Fails to Protect Witnesses 29.10.2026

Standard pitch-shifting boxes alter pitch and formants, but they leave distinct acoustic biometrics that speaker verification software can still recognize. Generative voice conversion offers an alternative by stripping identity-bearing acoustic information and resynthesizing speech with replacement embeddings in real time. Exploring research like Stream Voice Anon, latency limits, and the IARPA An...

Why Fast Speech Detectors Still Struggle with Onsets 29.10.2026

Voice activity detection models operate in milliseconds on a CPU, yet real-time conversational audio remains vulnerable to clipped speech onsets. Human hearing detects delays ranging from half a millisecond to tens of milliseconds, complicating systems like live voice changers. Frame-level benchmark scores mask onset jitter, while models like Silero, S4VAD, and FireRedVAD grapple with dataset anno...

Disguising Voices Without Losing the Emotion 29.10.2026

Documentaries are replacing traditional blurs and audio distortion with synthetic voices and digital avatars to protect vulnerable sources. Technical pipelines like F3VA and flow-matching try to separate vocal identity from emotional delivery, but entangled acoustics make selective anonymization difficult. Broadcast projects like Meshi Zahav and Hearing Their Voices demonstrate the power of post-p...

What Daylight Saving Actually Does to Health 29.10.2026

Moving clocks back and forth rearranges daily routines, but its measured effects on energy use, productivity, and health are surprisingly complex. A review spanning 157 studies across 36 countries challenges simple assumptions about seasonal time changes, highlighting modest increases in heart attack rates and traffic accidents in spring against reduced evening crime and lower autumn workplace inj...

Why People Reject the Official Clock 29.10.2026

National borders and administrative decrees frequently clash with the sun. Exploring how communities adapt when standardized time zones ignore local daylight, this discussion spans western China's five solar zones compressed into a single official hour, parallel clocks in Xinjiang, and the decades-long fight to place El Paso in Mountain Time. It examines how people manage morning routines, trade r...

Where Email Filtering Actually Happens 29.10.2026

Email filtering decisions come down to where the inspection occurs: pre-delivery MX gateways or post-delivery API tools. This breakdown explores the differences between gateway engines like Rspamd and SpamAssassin, examining asynchronous DNS lookups, Bayesian scoring, and neural network classification. It also covers modern post-delivery approaches, from enterprise behavioral tools to self-hosted...

Why Election Thresholds Discard Valid Ballots 29.10.2026

When thirty-eight party lists contest an election and only eleven clear the electoral bar, hundreds of thousands of valid votes end up unrepresented. This episode examines proposal tests for party registration, the mechanics of nationwide closed-list thresholds, and how discarded ballots alter seat math for qualifying parties. It compares registration practices across Australia, Mexico, and the Ne...

Why Israeli Election Turnout Means Opposing Things 29.10.2026

Israeli elections produced a record nineteen seats for Arab-majority parties, exposing a sharp division between foreign advocacy and domestic coalition building. While international advocates point to Arab voter turnout as living proof of a shared democracy, domestic campaigns frame that same political participation as a national security threat. The discussion explores how Hasbara shapes foreign...

Why Fast Backups Trigger Rate Limits 28.10.2026

API rate limits can stall automated backups even when accessing your own files. Looking at tools like rclone and the gws command line interface, this discussion explores how tiered quotas, token buckets, and hidden backend protections shape API behavior. Learn why method weights penalize recursive file listing, how shared project quotas mask individual activity, and why simply retrying with expone...

Filtering arXiv in the Age of Generative AI 28.10.2026

arXiv's monthly submissions doubled over two years to surpass forty thousand, driven by computer science and artificial intelligence. To manage the strain, the platform introduced a monthly submission cap of two papers per person. Filtering this volume requires navigating strict API limits, avoiding silent truncation in broad RSS feeds, and building digests that distinguish between institutional g...

Why Extra Voice Hours Settle Bilingual Speech 28.10.2026

Fine-tuning a massive speech model like Parakeet TDT on personal speech can yield dramatic gains with modest datasets, dropping code-switched word error rates from 13.7% to under 5%. How data maturity stages work, why repeated vocabulary adds acoustic resilience rather than redundant weight, and how models handle Hebrew-English code-switching without regressing on base English recognition are expl...

Backing Up Gmail with Got Your Back 28.10.2026

Downloading your own email locally turns out to require an entire Google Cloud project and an OAuth consent screen. This look at mailbox pipelines contrasts one-off Google Takeout archives with repeatable tools like Got Your Back and GAM. It covers how incremental EML backups work, how restore operations handle deleted mail without duplicating existing threads, the quirks of external OAuth test mo...

Why WhatsApp Business Backups Cannot Form an Archive 28.10.2026

Small businesses relying on client groups in WhatsApp Business often treat restore points as readable records. Examining standard cloud backups, linked devices, and tools like wtsexporter or iMazing reveals a fundamental gap: the platform offers restore points tied to active phone numbers, but no sanctioned method for automated, independent archiving. Preserving client correspondence and attachmen...

How Israel Actually Regulates Foreign Drugs 28.10.2026

Israel’s Health Ministry does not simply rubber-stamp decisions from the FDA or EMA. Regulators rely on unredacted assessment reports from six recognized authorities, requiring local dossiers to remain practically identical to foreign filings. The regulatory process, local packaging rules across four languages, and recent reliance pilot tracks shorten review times from hundreds of days down to sev...

Why Inhalers Never Combined Allergy Drugs 28.10.2026

Asthma inhalers have relied on the same core steroid and bronchodilator pairings for decades, rarely combining allergy-specific treatments into the same canister. Inhaled corticosteroids already suppress multiple inflammatory pathways, setting an unusually high bar for targeted additions like cromolyn or montelukast to show meaningful clinical benefits. While modern maintenance therapies expanded...

Why Common Colds Cause Asthma Flare-Ups 28.10.2026

Viral infections frequently trigger asthma exacerbations, but the molecular mechanisms reveal an unexpected asymmetry. Rhinovirus can drive airway epithelial cells into severe inflammatory overdrive while simultaneously hobbling their ability to produce antiviral interferons and clear the pathogen. This breakdown covers what actually happens inside inflamed lungs, why air trapping makes exhaling d...

Choosing Ergonomic Inputs for Working in Bed 28.10.2026

Working from bed exposes the hidden floor-space requirements of standard ergonomic pointing devices. Tackling an eighteen-month email backlog demands reliable bulk selection without strained reaching. This review examines practical bed-friendly computing setups, comparing firm lap trays, thumb trackballs like the Logitech MX Ergo S and ProtoArc EM01 NL, integrated trackpad boards like the Logitech...

Can Diplomacy Repair Economic Isolation? 28.10.2026

Fragmented European trade restrictions, post-election deadlock, and conflicting statements from officials have created unprecedented commercial friction. An examination of settlement trade bans in Spain and the Netherlands reveals how piecemeal rules drive corporate overcompliance long before formal banking sanctions arrive. Resolving these disputes requires distinguishing policy-driven trade limi...

Why Israeli Youth Moved Right 28.10.2026

The assumption that younger generations naturally tilt liberal breaks down in Israeli politics, where up to three-quarters of new voters identify with the right. An examination of recent election polling, cohort research by Liran Harsgor, and insights from Alon Yakter explores what drives this political shift. It unpacks the roles of physical separation, changing media habits, and differential fer...

Why Early Israeli Election Tallies Skew Right 28.10.2026

Early election counts can shift dramatically overnight, but not because voters changed their minds after polls closed. The national tally updates as complete polling stations report, meaning initial figures reflect small, fast-counting locations rather than a representative cross-section. The breakdown unpacks station-level reconciliation, transit delays to regional election committees, and the me...

Lyt til podcasten My Weird Prompts i Replaio

Radio og podcasts i én app - gratis og uden tilmelding. Installer i dag, og gå ikke glip af lanceringen

Hent den på Google Play

Replaio er ikke podcastudgiver; programmernes navne, coverbilleder og lyd tilhører deres ophavsmænd og distribueres via offentlige RSS-feeds.