#5438: Why Linguists Left the AI Room

Large language models grew out of linguistics — so why aren't linguists in the room where they're built?

transformer-architecturetraining-dataai-evaluation

#5437: Hunting the Overused Phrase in Our Own Scripts

There's no tool for finding an LLM's verbal tics — you have to build one. Here's how keyness analysis works.

ai-evaluationtraining-datahallucinations

#5436: Small Models as Rewriters, Not Writers

Why "don't say X" prompts backfire, and how a tiny grammar-constrained model can scrub a script without breaking its grammar.

prompt-engineeringhallucinationslocal-ai

#5435: When Your TTS Model Eats the Numbers

Numbers, dates, and acronyms break text-to-speech in specific, documented ways. Here's where normalization lives — and why it depends on your model.

text-to-speechspeech-recognitionai-inference

#5434: When Local AI Tools Phone Home

Local AI tools shouldn't need the network. So why is your on-device writing assistant uploading 40MB at 3am?

cyberattacksdigital-privacylocal-ai

#5433: Thermostats, Thermal Mass, and the Buildings That Fight Back

A thermostat is a sensor, a set point, and a switch — until thermal mass, bad models, and sustainability mandates get involved.

therapy-and-counseling

#5432: What Hospice Workers See at the Deathbed

A UK survey found 97% of palliative care staff witnessed deathbed phenomena. What they see — and why 92% never makes it into the chart.

stress-and-traumarelationshipstherapy-and-counseling

#5431: The Other Half of Hugging Face: Why BERT Still Out-Downloads Llama

Encoder models pull over a billion downloads a month. Decoder models pull 397 million. The AI conversation and the download counter are describing ...

transformer-architecturefine-tuningai-evaluation

#5430: Two Boxes: ASR and the Text Fixer Behind It

Punctuation, casing, ITN, disfluency — the four-job layer between raw ASR output and text you can actually read.

speech-recognitiontext-to-speechtransformer-architecture

#5429: Bathtime Safety: What the Data Actually Says

Two-thirds of child drownings inside the home happen in the bathtub. Here's what the research says about temperature, supervision, and the towel.

baby-gearchild-safetypublic-health

#5428: Wills, Superstition, and the One Event With 100% Probability

Daniel is a prepper who's terrified of writing a will. Turns out the law, the psychology, and the Israeli defaults all agree he shouldn't be.

courts-and-judiciaryrelationshipsstress-and-trauma

#5427: Why Atheists Still Knock on Wood

Religion and superstition aren't the same thing — and the data says they might even pull in opposite directions.

cognitive-biasesrelationshipshabits-and-behavior-change

#5426: AI Wrote His Yom Kippur Prayer

A man skipped synagogue, opened ChatGPT instead, and got a prayer that knew his life. Is AI-generated prayer real prayer?

ai-alignmentai-evaluationcontext-and-memory

#5425: Preview Branches Without the Branch Dance

Daniel dreads the branch-switching dance. Turns out Vercel's promote-to-production makes most of those steps not exist.

devopsai-coding-assistantsweb-development

#5424: ThinkPad E15: Swollen Battery or Bad Keyboard?

A dead arrow key and a jammed trackpad on a two-year-old ThinkPad E15 might be one problem, not three.

batteries-and-powerpc-hardwareconsumer-protection

#5423: Underground Burial in Jerusalem: 24,000 Graves, 3 Elevators

Jerusalem ran out of ground for its dead — so it built a multi-storey car park where the parking spaces are graves.

jerusalemjewish-historyisrael

#5422: Switching Android Keyboards Without the Tap Dance

One listener wants a one-tap jump between his Parakeet voice keyboard and his typing keyboard. Android says no — unless you know the trick.

hallucinationsspeech-recognitionsmartphones

#5421: Why Secular Israelis Still Fast on Yom Kippur

A majority of secular Israeli Jews fast on Yom Kippur. What the data say about how faith gets handed down — and how observance becomes a weapon.

jewish-holidaysjewish-lawisrael-defense-forces

#5420: How Keyboard Middleware Actually Works

A keyboard isn't one thing — it's a stack. Here's how input methods, composition buffers, and hooks really work on Linux, Android, and Windows.

smartphonesspeech-recognitionai-inference

#5419: The Attention Budget in Your Pocket

Why phone dictation runs out of room, and how bounded attention windows buy you punctuation without blowing your memory budget.

speech-recognitionai-inferencelocal-ai

#5418: Why Dehydration, Not Hunger, Is the Real Risk of a 25-Hour Fast

Twenty-five hours without water is a different problem than twenty-five hours without food. Here's the physiology of dehydration — and how to prepare.

public-healthnutrition-and-dietexercise-and-fitness

#5417: Can You Version-Control Your Android Apps?

Android Developer Verification enforcement just started. Can you define your installed app list as code — and what breaks if you try?

open-sourcesmartphonescyberattacks

#5416: Xet, Buckets, and Auto-Pulling Model Weights

Hugging Face swapped its storage backend to Xet with barely a ripple. So why can't a bucket auto-pull new upstream weights?

training-dataopen-sourceopen-source

#5415: Inside F-Droid: Repos, Front Ends, and Sideloading Safely

How F-Droid's repository model works, why it has five front ends, and what actually keeps sideloaded APKs safe.

open-sourcesmartphonescyberattacks

#5414: Inventing Anna: The Real Tradecraft Behind the Con

How Anna Delvey ran a four-year con on Word docs, AOL addresses, and a supermarket burner phone — and what Netflix got right.

cyberattacksfraud-and-scamsfilm-and-television

#5413: Who Actually Picks Your In-Flight Movie?

Eight in ten passengers use seatback screens — but at Southwest, one person picks everything. Inside the tiny teams behind airline entertainment.

aviation-safetytourismmarket-competition

#5412: Editing vs. Note-Taking for AI Fine-Tunes

Hand-editing a model's output gives three training signals at once. Writing notes gives one — and a weaker one at that.

fine-tuningtraining-dataai-evaluation

#5411: Fine-Tuning at 4-Bit vs 16-Bit: What It Really Costs

QLoRA cuts fine-tuning VRAM 15x and cost up to 85% — but you pay in training time, quality, and safety alignment.

fine-tuninggpusai-inference

#5410: Adapters: 102KB That Reshapes a 403GB Model

A 102KB adapter file changes how a 403GB base model behaves — without ever merging into it. Here's how model adapters actually work.

fine-tuninglocal-aitransformer-architecture

#5409: What Makes a Model "Agentic"? Atria Dawn Preview

A 744B-parameter model drops quietly on Hugging Face. Is "agentic" a real model category, or just a deployment pattern?

ai-evaluationfine-tuningai-inference

#5408: When Small NLP Models Beat the LLM

Feature extraction, fill-mask, token classification — the classic NLP tasks still have a job. Here's when a small model beats a frontier API.

fine-tuningai-inferenceai-evaluation

#5407: Hemmingway-1 and the War on Waffle

A 27B model promises answers without the preamble. Its benchmark is homegrown — and the behavior it targets has a paper trail.

ai-evaluationfine-tuningai-alignment

#5405: Omarchy: The Linux Distro Built for AI Agents

Omarchy treats AI agents as users of the OS itself — every setting a command, every config a text file. Here's how it works and where it breaks.

linuxai-coding-assistantsopen-source