Advertise on podcast: My Weird Prompts

CENO gets covered as a border-tunneling story, but the real design is closer to "does anyone on my street already have this page." This episode walks the mechanics end to end: the Ouinet client proxy, the signing injector run by eQualitie, bridges that relay ciphertext they can't read, and seeders who volunteer by accident just by browsing. We trace a single page request through the race between direct, injector, and peer cache routes, explain how the BitTorrent mainline DHT and uTP transport carry signed pages between phones, and answer the question everyone asks: how many intermediaries can a chain hold? The answer is one bridge, and the reason isn't security — it's that every extra hop is another thing that can fail. Episode 6041 — myweirdprompts.com/n/6041 (permanent short link: myweirdprompts.com/271038)

Claude Code now lets subagents spin up their own subagents — so what happens when the delegation stack goes three, four, five levels deep? We trace the real history of spawn-depth defaults (five layers, then one, then three), the environment variable that controls it, and the striking convergence across Gemini CLI, Cursor, Google's ADK 2.0, Anthropic's Managed Agents, the OpenAI Agents SDK, and LangGraph. Then we get to the uncomfortable part: the published literature on recursive delegation stops at depth three. We dig into the ADaPT paper from NAACL Findings 2024, MIT's Recursive Language Models, and Anthropic's multi-agent research numbers — fifteen times the tokens, a single global budget cap, and no per-branch containment. If you've wondered whether A2A is running under the hood, the answer is no — and what's there instead is more interesting. Episode 6040 — myweirdprompts.com/n/6040 (permanent short link: myweirdprompts.com/045336)

Only two countries are at the AI frontier right now — and everyone else is training models for entirely different reasons. This episode works through the real numbers: Mistral's trillion-parameter preview scoring 38 against a frontier of 58, the UAE's fully open K2 Horizon, Korea's government-run elimination rounds, India's from-scratch Sarvam models, and more. Along the way we untangle why older index scores make sovereign models look far stronger than they are, and why "sovereign AI" means something different in Seoul, Abu Dhabi, Paris, and New Delhi. If you've ever wondered who's actually close to the frontier and who's building for language coverage, data residency, or procurement rules instead, this is the episode. Episode 6039 — myweirdprompts.com/n/6039 (permanent short link: myweirdprompts.com/460947)

DeepSeek V4.1 Flash writes this show's scripts — but is it still the right model to pay for? This episode walks through eight Chinese labs with flagships a Western developer can actually reach, from Alibaba's Qwen3.8 Max to Moonshot's Kimi K3, Zhipu's GLM-5.3, Xiaomi's MiMo, MiniMax, StepFun, Tencent and Ant Group. Along the way: specs nobody has reconciled, a headline price that turns out to be one fp4 host, peak-hour billing windows, and the verbosity tax that doubles your real cost per task. Then the Western frontier enters as a yardstick — and the open-weight gap turns out to be about twelve points for roughly forty times the price. Episode 6038 — myweirdprompts.com/n/6038 (permanent short link: myweirdprompts.com/616052)

Google's Gemini 4 Argon is positioned as a frontier reasoning model, but the model card leaves out architecture, parameter count, and licensing, and the page quotes no pricing at all. We walk through what's actually documented: a 1,048,576-token unified input/output limit, low/medium/high thinking levels, implicit caching only, and a wide tool surface including function calling, code execution, Search grounding, and a preview computer-use mode. Then we separate Google's nineteen-row benchmark table from independent testing — Artificial Analysis at 53 on the Intelligence Index, Vals AI ranking it first of 49 models — and note where the losses cluster: terminal-based agentic coding and computer use. We also cover the staged rollout through the Fairwind Program and the launch discount that changes the cost-per-task math. Episode 6037 — myweirdprompts.com/n/6037 (permanent short link: myweirdprompts.com/889907)

OpenAI's GPT-6 Astra ships with a 1.05 million-token context window, ten native tools including computer use and MCP, and a five-level reasoning effort dial — but also no fine-tuning, a long-context pricing cliff, and benchmark numbers that shift depending on which harness ran them. This model spotlight walks through the spec sheet, the pricing math, and the contested ARC-AGI-3 results, plus the safety context: two training pauses in three months and reported departures from the safety team. If you're building on frontier models, this is the tradeoff map you need before you commit. Episode 6036 — myweirdprompts.com/n/6036 (permanent short link: myweirdprompts.com/110245)

Daniel asked a simple question: who else is out there making AI-hosted podcasts, and what have they learned? The answer turned out to be stranger than expected — the format is far rarer than it looks. This episode surveys the closest matches in the world, from a two-AI-host show that ran eleven episodes and went silent, to a developer whose serverless pipeline produced 730 daily episodes over two years until he got emotionally attached to a robotic mispronunciation. Along the way: a human host whose AI callers have "secret wants," a fully automatic Hacker News recap that stopped without a word, and an open-source Chinese-language project that's still ticking. The recurring pattern? The shows that survive are the ones where the maker refuses to automate one small, personal step. Episode 6035 — myweirdprompts.com/n/6035 (permanent short link: myweirdprompts.com/840344)

Anthropic released Claude Opus 5.5 on September 22, 2026, and the reception was loud. In this Model Spotlight episode, we profile what the lab actually shipped: a cheaper price across every line than Opus 5, a hard change that means thinking mode can no longer be switched off, and a model card that conspicuously omits parameter count, architecture, and context window. We walk through the benchmark table, the independent verification from Artificial Analysis, the pricing snapshot as of mid-October, and the anti-distillation safeguard Anthropic calls preserved thinking. If you're doing capacity planning on a frontier model, the gaps in this card matter as much as the numbers. Episode 6034 — myweirdprompts.com/n/6034 (permanent short link: myweirdprompts.com/321817)

Progressive skills discovery is treated as solved — until your skill listing quietly drops descriptions and skills stop firing. This episode digs into how Anthropic's three-tier progressive disclosure actually works, why Claude Code truncates descriptions for your least-used skills first, and how Codex handles the same problem with different numbers. We survey the open-source landscape for auto-syncing with marketplace updates and semantic search over skill definitions, including why the one project that hit both is now abandoned. If you're maintaining a self-hosted skill catalogue behind an MCP gateway, this is the episode about whether the architecture you built has a future. Episode 6033 — myweirdprompts.com/n/6033 (permanent short link: myweirdprompts.com/345391)

If your use case is narrow enough to find training data for, blending multiple datasets into one fine-tune run can beat merging finished checkpoints. This episode walks through the full pipeline: why uniform or size-proportional sampling is usually suboptimal, the formatting tax that scales with every source you add, and why deduplication isn't optional — memorization scales superlinearly with repetition, and one dedup study saw up to ~20% perplexity improvement while cutting runtime nearly 28%. Then the tools: Hugging Face's concatenate and interleave, datamix, Oumi, LLMTune, and NeMo's code-switched data scripts — and why there's no single product that owns all four stages. Key sources: • mergekit repository (Arcee AI) — https://github.com/arcee-ai/mergekit • Hugging Face `datasets` — https://huggingface.co/docs/datasets/process • Hugging Face `datasets` combine.py source — https://github.com/huggingface/datasets/blob/main/src/datasets/combine.py Episode 6032 — myweirdprompts.com/n/6032 (permanent short link: myweirdprompts.com/882072)

Daniel and Hannah are both self-employed, raising a fifteen-month-old in a small Jerusalem rental with no family nearby. Daycare runs seven to three, six days a week — and they're still exhausted. This episode digs into why. Time-use research shows it's the shape of the hours, not the count: fragmented work time predicts time pressure independently of total hours. The mental load literature explains what never gets handed off at daycare pickup. The cooperative-breeding anthropology explains why the isolated parental pair is the historical outlier, not the norm. And the ADHD angle flips the usual advice — protect the block instead of getting better at being interrupted. Plus: what's actually temporary (the recovery) versus what's load-bearing (the structure), and five concrete changes ranked by impact. Key sources: • Sear, *Phil. Trans. R. Soc. B*, 2021 — https://pmc.ncbi.nlm.nih.gov/articles/PMC8090810/ • Reich-Stiebert et al., *Sex Roles*, 2023 — https://pmc.ncbi.nlm.nih.gov/articles/PMC10148620/ • Loop Earplugs — https://www.loopearplugs.com/pages/earplugs-for-parenting Episode 6031 — myweirdprompts.com/n/6031 (permanent short link: myweirdprompts.com/216874)

If you were building a team to train a model — not at OpenAI or DeepMind, but at a small lab or enterprise — who would you hire? The answer is stranger than the job postings suggest. At the engineering layer, titles are collapsing: Anthropic lists everyone as a Member of Technical Staff because the research/engineering boundary stopped being useful. At the data layer, they're multiplying: dataset engineers, licensing attorneys, annotation specialists. This episode walks through what each role actually does all day, which jobs fold into one person when you're small, and what happens when regulation — not fashion — drives demand for a career path that barely existed five years ago. Key sources: • The AI Roles Continuum, arXiv (2026) — https://arxiv.org/html/2601.06087v1 • Google, Assembling an ML team — https://developers.google.com/machine-learning/managing-ml-projects/team • Anthropic, Research Engineer, Pretraining Scaling (2025-09-30) — https://job-boards.greenhouse.io/anthropic/jobs/4938432008 Episode 6030 — myweirdprompts.com/n/6030 (permanent short link: myweirdprompts.com/355716)

Daniel asks whether the whole model landscape is converging on one universal model you configure for anything — or whether specialization stays fundamental. Two test cases: reasoning, which started as its own model class and now looks like a knob, and instruction tuning plus structured output, where the base/instruct/chat menu seems to be collapsing into one steerable model. But there's a sharper question underneath: are we watching architecture consolidate, or just packaging? OpenAI's o-series folded into GPT-5's routed system, DeepSeek actually merged chat and reasoner into one set of weights, Qwen walked its hybrid back to separate models, and Anthropic went from an effort parameter to adaptive thinking with no manual control at all. Three roads, one signpost — and the underlying differences didn't shrink in any of them. Key sources: • OpenAI, 2026-04-23 — https://openai.com/index/introducing-gpt-5-5/ • Anthropic, 2025-11-24 — https://www.anthropic.com/news/claude-opus-4-5 • Anthropic — https://www.anthropic.com/news/claude-opus-4-6 Episode 6029 — myweirdprompts.com/n/6029 (permanent short link: myweirdprompts.com/754710)

A year ago, the prediction was that dedicated ASR models would go extinct — why build a transcription model and a second model to clean up its output when a multimodal LLM could do both from raw audio? Two years of evidence says otherwise. The top of the Open ASR Leaderboard is still owned by Conformer-encoder-plus-LLM-decoder systems on accuracy, but a 0.6B TDT model sits four tenths of a point behind at roughly eight times the speed, and a new third layer has appeared: specialized post-processing models like Superwhisper's S1-mini that turn raw transcripts into formatted text. This episode works through the three-layer stack, why the latency gap is an architectural fork rather than a maturity gap, why word error rate is under active attack as a metric, and why no benchmark yet compares audio-to-formatted-text against ASR-plus-normalization head to head. Key sources: • Open ASR Leaderboard paper (v2, 2025-10-08 data) — https://arxiv.org/html/2510.06961v2 • NVIDIA Parakeet TDT 0.6B v3 model card — https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3 • S1-mini model card — https://huggingface.co/superwhisper/s1-mini Episode 6028 — myweirdprompts.com/n/6028 (permanent short link: myweirdprompts.com/351585)

Daniel in Jerusalem has two NVIDIA Parakeet fine-tunes — one for Hebrew-English code-switching, one for niche technical vocabulary — and wants to know if he can merge them like LEGO bricks instead of retraining on a combined dataset. It's a question with a combinatorial edge: hospitals, legal teams, and multilingual clinics all want pairings nobody will ever build dedicated models for. We walk through what merging actually does to weights — task vectors, weight averaging, TIES-Merging, DARE, and LoRA adapter merging — and ask whether transcription's demand for exactness makes unified training the only honest path. Along the way: why word error rate can hide the exact failures you fine-tuned to fix, and why fine-tunes from the same base checkpoint come pre-aligned. Key sources: • TIES-Merging: Resolving Interference When Merging Models — https://arxiv.org/html/2306.01708v2 • An Introduction to Model Merging for LLMs — https://developer.nvidia.com/blog/an-introduction-to-model-merging-for-llms/ • NeMo Speech Fine-Tuning docs — https://docs.nvidia.com/nemo/speech/latest/asr/fine_tuning.html Episode 6027 — myweirdprompts.com/n/6027 (permanent short link: myweirdprompts.com/991040)