Local AI
394 episodes · Page 5 of 17
#6191: What Actually Happens During Fine-Tuning
Sixteen bytes per parameter in, two out — a walk through the forward pass, backward pass, and optimizer that turn a dataset into new weights.
#6190: When Your AI API Quietly Changes Under You
Same API name, different model behind it. How to tell when the thing you tested stops being the thing you're using.
#6188: Every Way Linux Relieves Memory Pressure
Zram, zswap, swappiness, and the OOM killers — the full ladder Linux climbs when RAM runs out, and what to actually run.
#6185: Why AI Labs Sell a Tier Above the Flagship
Fable, Deep Think, SuperGrok Heavy — what the $300/month top tier actually does that the flagship can't.
#6182: Running a Private APT Repo for Family Machines
Daniel wants his Debian packages on family machines without becoming their unpaid update service. Here's the modern way to do it.
#6177: How Voice Cloning Works From Six Seconds of Audio
Six seconds of reference audio is all Chatterbox needs. The real work happened in the half-million hours of training before you showed up.
#6176: Do Audio Models Hear How You Said It?
A benchmark called LISTEN found audio models score near chance when emotion lives in the voice, not the words. What does that mean for real calls?
#6175: Video Models Can't Tell Time: What Motion Benchmarks Reveal
Frontier models score 48% on tasks where humans hit 98% — the gap is motion, speed, and sequence, not scene description.
#6172: Recipes vs. Regeneration: AI Photo Editing in 2026
Two ways AI edits photos now: models that write a readable recipe of slider values, and models that just repaint the whole frame.
#6171: AI Audio Restoration: What Actually Ships in 2026
Noise removal and bandwidth extension are shipping products today. Matching a clip to a target EQ is still mostly a research prototype.
#6170: One Endpoint, Many Boxes: Local LLM Routing
A LiteLLM-style proxy can fail over between your own GPU boxes and the cloud — but it can't see your VRAM.
#6168: Hebrish: Hebrew-Aware Dictation That Actually Works
Two new speech models fix English-Hebrew code-switching dictation — and the whole training run cost about fifty cents of GPU time.
#6163: Forking a Keyboard for One Voice Feature
Daniel built a Hebrew-English dictation engine. Now he wants to fuse it into someone else's keyboard — without inheriting its entire future.
#6162: Why Dictated Words Rewrite Themselves
Hold-to-talk dictation shows dimmed text that silently changes when you release the key. Here's the architecture behind that rewrite.
#6160: Where Do Tiny Speech Rules Live? A Ladder From Regex to LoRA
A regex notebook, a WFST grammar, or a fine-tune? Where do tiny speech-rewrite rules actually belong in a 600M-parameter model?
#6157: Where to Sit When You're Fine-Tuning Your Own Voice
Your phone's built-in mic is the point — the setup around it is what ruins the takes. Rooms, remotes, and the app setting that deletes your data.
#6148: The Tiny Hebrew Models Behind the Show's Voice
Dicta's 300M-parameter nikudizer, Phonikud's adaptors, and the ktiv male model hiding inside a vowel adder.
#6147: Hebrew AI: Ears, Mouth, and Brain
Open Hebrew speech recognition beats the paid services. Hebrew text-to-speech is a different story — and the language models are a third.
#6139: Which AI Writes Our Scripts Best? We Tested Five
Five models, three scripts each, one rubric. Sol won on length, Grok on voice — and the runner-up title hides a catch.
#6138: Hebrish on iPhone: MLX, Core ML, or Wait?
Hebrish spells makolet right on Android. Getting that Hebrew-English dictation onto an iPhone means picking between MLX, Core ML, and sherpa-onnx.
#6135: How Many Runs Does Your AI Eval Actually Need?
Reruns shrink one kind of noise and leave another untouched — so where should your eval budget really go?
#6134: Cheap Tokens, Expensive Scripts
A model with worse list prices can cost less per script — if the cache hits. The bill hides in tokens, not rates.
#6132: How to Run a Model Eval on a Budget
Daniel is teaching himself to run evals on a real budget. Successive halving, paired comparisons, and why three clean scripts prove almost nothing.
#6127: Haiku 5.5 at 10 Cents: What the Cheap Model Actually Does
Haiku 5.5 costs 10 cents per million input tokens. We break down where it holds up, where it fails, and whether it can write this show.