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 19, 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 Transformers Process Images and Video 27.10.2026

Transformers originally built for language now power visual perception and image generation. Replacing convolutional networks with self-attention requires turning continuous pixels into discrete vector sequences, which Vision Transformers achieve by treating raw image patches like text tokens. Exploring architectures like ViT and Diffusion Transformers (DiT) shows how swapping inductive visual bia...

Why Model APIs Lack Dedicated Few-Shot Fields 27.10.2026

General model APIs treat few-shot demonstrations as raw text or chat messages rather than structured, versionable task artifacts. While specialized models like iAmTime and LoRA Recycle train directly on explicit demonstration structures, frontier LLM platforms rely on unstructured prompts, XML tags, or massive context windows. An examination of in-context learning shows why adding an examples para...

Why DeepMind Systems Live Outside the Gemini API 27.10.2026

Google DeepMind regularly announces breakthrough systems that never appear as simple endpoints in Google AI Studio or the Gemini API. Exploring systems like AlphaEvolve, Co-Scientist, Genie, and AlphaFold reveals why they are kept separate. The restrictions stem from serving costs, workflow requirements, and commercial strategy rather than mere secrecy, requiring specialized cloud setups, research...

How Google DeepMind Absorbed the Gemini App 27.10.2026

Google DeepMind is often treated as an academic research enclave, but recent reorganisations have turned it into Google's core consumer AI division. Tracing the trajectory from the 2023 merger with Google Brain to the October 2024 shift of the Gemini app team under DeepMind leadership, this episode unpacks the corporate reality behind the branding. It examines UK statutory accounts, leadership cha...

Why Exact Matching Outlasts the Index 27.10.2026

Searching without building a massive index remains surprisingly durable. While modern developer tools often rely on complex indexing, streaming Unix utilities like grep, find, and locate still handle immense workloads by balancing query-time scanning against stale precomputed paths. Exploring Ken Thompson's original 1973 assembly implementation, the Boyer-Moore skipping algorithms behind modern ma...

Training Custom Rerankers for Podcast Search 27.10.2026

Building custom search across a dense podcast archive requires more than off-the-shelf vector retrieval. The real engineering work happens during candidate reranking and data curation. This look at search architecture explores chunking transcripts, navigating the tradeoff between recall ceilings and inference costs, mining hard negatives without accidentally penalizing nuanced matches, and using q...

Why Terminal Fuzzy Finders Forgive Different Mistakes 27.10.2026

Approximate string matching lives between strict regular expressions and semantic retrieval, but different search tools play by entirely different rules. While web search engines rely on edit distances to forgive typos and transpositions, terminal finders like fzf require strict ordered subsequences. This exploration covers how algorithms from Levenshtein, Damerau, Bitap, and Smith-Waterman comput...

Why Semantic Desktop Search Struggles to See Files 27.10.2026

Semantic vector search is finally arriving on operating systems, but finding files locally remains surprisingly tricky. Operating system updates to Apple Spotlight and Windows Search now bring local embedding models to personal computers, yet application boundaries, missing OCR, hardware gates, and cloud placeholders create blind spots. Examining native implementations alongside independent tools...

Blind Testing AI Video Models for Architects 27.10.2026

Evaluating AI video models requires more than skimming glossy vendor demos, especially when an architectural project cannot tolerate shifting walls or warping window mullions. Evaluating tools like GenMedia, media-gen-arena, and GenMediaBenchmark reveals the practical mechanics of running internal blind pairwise comparisons. Rigorous testing demands tracking settings, managing per-clip generation...

Why Fine-Tuning Whisper Won't Stop Duplicated Text 27.10.2026

Speech recognition errors are often blamed on model weights when the real culprit is inference orchestration. Fine-tuning an acoustic model will not fix duplicated phrases caused by chunk merging, fallback temperature schedules, or silence heuristics. By examining Whisper's DecodingOptions, beam search trade-offs, and WhisperX's choice to disable conditioning on previous text, this discussion maps...

Building an AI Rubric to Rank Israeli Parties 27.10.2026

A voter's 42-minute personal recording turns into an explicit voting rubric for Israel's twenty-sixth Knesset election. Using Claude research agents to score platforms against prioritized issues like tenant protections, cost of living, and public services rather than standard headline debates, the experiment yielded surprising party rankings. An examination of the findings reveals a striking polit...

Choosing the Right Audio Annotation Tool 27.10.2026

Turning raw environmental recordings into machine learning datasets requires choosing between lightweight inspection tools and heavyweight crowd platforms. From offline desktop staples like ELAN and Praat to self-hosted frameworks like Label Studio, Potato, and bioacoustic-focused Whombat, each tool handles timeline events differently. The discussion covers tier management for overlapping sounds,...

Finding Recurring Urban Sounds With Four Examples 27.10.2026

Training an audio detector does not always require hundreds of manual annotations. Borrowing methods from bioacoustic few-shot sound event detection, a system can use pretrained audio neural networks to locate recurring noises like train horns from just four or five distinct examples. The discussion covers model-assisted labeling workflows, using prototypical networks on top of PANNs embeddings, a...

Why Dictation Models Steal Other People's Voices 27.10.2026

Speech recognition models can transcribe every word accurately while capturing the wrong person's conversation. Fine-tuning a recognizer on a single voice does not teach it to ignore clean, loud background speech like train platform announcements. Addressing issues in systems like Parakeet requires target-speaker conditioning, diarization-dependent masks, or enrollment embeddings rather than raw a...

Teaching Dictation Models to Ignore Other Voices 27.10.2026

Standard speech recognition will faithfully transcribe clear background announcements into a personal document. Target-speaker automatic speech recognition attempts to fix this by enrolling an authorized user's voice and conditioning the transcription on that acoustic identity. A look at how speaker embeddings, attention conditioning, and architectures like Speaker-Querying Whisper handle interfer...

Why Speech Recognizers Cut Off Hesitant Speakers 27.10.2026

Speech recognition systems routinely cut callers off mid-thought, mistaking a hesitant pause for a finished turn. The mechanics behind endpointing reveal why accurate vocabulary models still fail in emergency dispatch and accessibility tools. Looking at pause frequency, duration, and repair structure across spontaneous speech, older adults, and clinical cohorts shows that human hesitation follows...

When Splitting Serverless Apps Earns Its Keep 27.10.2026

Architectural boundaries often reflect deployment schedules rather than execution limits. On platforms like Modal, grouping web endpoints, background cron jobs, and heavy GPU workers into a single app provides atomic deployments without sharing containers, scaling budgets, or container images. Splitting these workloads into separate apps isolates blast radiuses and decouples release cadences, but...

Why Tool Calls Break Across Coding Models 27.10.2026

Moving coding setups from Claude Code to environments like OpenCode using Kimi models reveals hidden incompatibilities beneath simple dropdown switches. Tool schemas, Anthropic's user-role tool results, and provider-specific chat templates handle calls entirely differently. Inspecting how native tokenizer tokens and inference parsers package structured function calls explains why an agent might de...

What Dental Hygiene Teaches Us About Benchmarks 27.10.2026

Assessment problems in AI evaluation mirror longstanding challenges in healthcare education. Dr. Ann McCann spent decades redesigning how dental hygiene schools verify real-world competence, discovering that most program directors relied on unvalidated tests. Rebuilding evaluation around observable performance tasks rather than multiple-choice knowledge offers a practical blueprint for benchmarkin...

Predicting Random Signals When Statistics Drift 27.10.2026

Standard statistical forecasting leans heavily on stationarity, assuming signals keep stable averages over time. Real systems drift and shift, prompting mathematicians like Marc Mehlman to explore harmonizable processes. Building on M. M. Rao's dilation theorem and Wold's decomposition, Mehlman's research extended fundamental moving average representations and optimal linear prediction formulas to...

Surplus Vote Math in the Knesset 27.10.2026

Choosing between Yashar and BeYachad reveals how electoral mechanics complicate voter intent in Israel's proportional system. A ballot elects a closed list rather than a prime minister, leaving negotiations over the top post to post-election bargaining. The breakdown explores how the Bader-Ofer allocation rules, nationwide single constituencies, and separate surplus agreements between paired parti...

Polymarket's Final Odds for Israel's 2026 Vote 26.10.2026

Prediction market prices on the eve of Israel's election suggest a fractured outcome. With Yashar leading Likud in the polls and markets, Gadi Eisenkot stands as the favorite for prime minister at 55.9 percent, but traders place the odds of any declared bloc securing a 61-seat majority at just 16.5 percent. A breakdown of late trading volumes, coalition combinations, parliament deadlock risks, and...

Why Hermes and Pi Top the OpenRouter Coding Table 26.10.2026

Token rankings place Nous Research's Hermes Agent and Mario Zechner's pi above Claude Code on OpenRouter's coding leaderboard. The leaderboard's measurement quirks explain part of the gap, but the real divide is architectural: Hermes builds an extensive persistent apparatus with strict character-limited memory files, while pi opts for an aggressive four-tool core with branched session trees. The a...

What Happens Between Model Checkpoints 26.10.2026

Public releases often treat an AI model as a completed artifact, but the real gains frequently happen long after pretraining stops. Exploring checkpoint transitions reveals how labs like Ai2 and Meta rely on continuous post-training routines. From extending reinforcement learning runs on existing datasets to orchestrating repeated rounds of rejection sampling and direct preference optimization, mi...

Why Model Names Conceal Changes Behind the API 26.10.2026

API model identifiers frequently hide underlying updates, version changes, and system adjustments behind a single static label. While managed aliases offer automatic improvements and convenience, they can obscure the link between published evaluation cards and the live service answering requests. Research on silent updates reveals how routing, classifier tweaks, checkpoint selection, and model mer...

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.