Heretik

Fringe Tech 28 Aug 2026

Prompt Injection Works Because the Model Can't Tell Who Is Talking

New research traces prompt injection to a single mechanism: LLMs decide who is speaking from writing style, not from role tags. Forged reasoning takes attacks from near-zero to 60% success. Remove the style, and it collapses to 10%.

5 min read

Fringe Tech 28 Aug 2026

NVIDIA Is Reportedly Buying Hugging Face. What Would It Mean for Local AI?

One anonymous source says $12.9 billion. Both companies are silent. Here is what is actually known, what the communities fear and hope, and the questions worth asking before anyone predicts anything.

6 min read

Fringe Tech 28 Aug 2026

Running Qwen3.8-Flash-Next at Full 262K Context on a 128GB MacBook

Benchmarks for Qwen's new hybrid-attention MoE at its native 262K context on a 128GB M5 Max: the architecture that makes it fit, the day-0 recipe, and a depth sweep from 0 to 262K tokens.

7 min read

Essays 26 Aug 2026

Judas: Sold Out at Every Turn

A documentary reconstruction of Judas Iscariot and the Codex Tchacos, laid out as evidence and left for the reader to weigh.

46 min read

Fringe Tech 25 Aug 2026

Reading a DSC Alarm Panel's Keybus with a $7 ESP32

A late-1990s DSC PowerSeries panel already knows every door and window in the building. Two resistor dividers and a XIAO ESP32-S3 read all of it into Home Assistant and Grafana, read only.

5 min read

Rant 16 Aug 2026

Fine Art America, Pixels.com, and the Corvette Puzzle That Never Shipped

Over sixty dollars for a two-hundred-piece cardboard puzzle, ordered two full weeks ahead, never shipped. Four days for a first reply, a week more for the refund, and a complaint record that was public the whole time. A consumer autopsy of Fine Art America and Pixels.com.

3 min read

Philosophy 14 Aug 2026

Building a Mind on Open Questions

Preferring questions that cannot be answered over answers that cannot be questioned is load-bearing for a sound psyche. On intolerance of uncertainty, beliefs welded to the self, Hoffer's engineered certitude, and open questions as the joints a mind moves on.

8 min read

Philosophy 24 Jul 2026

Two Poisons Make Salt

Atoms bond because of what they lack. The gap you have been hiding is the only place another person can attach. On complements, activation energy, and being the catalyst for somebody else.

10 min read

Fringe Tech 24 Jul 2026

Laguna-XS runs beautifully on my 5090, and I am keeping the model everyone calls outdated

A newer, well-recommended coding model fits my 32GB GPU at max quality and runs faster than the one I keep. I kept the old one anyway, and the numbers say I was right to.

7 min read

Fringe Tech 22 Jul 2026

Qwen3.6-27B Scores Higher on BFCL, Ties on My Workload, and Runs 4.9x Slower

Qwen3.6-27B scored higher on BFCL. Both models scored 58/59 on the real workload. One answers in 0.87s, the other in 4.27s.

10 min read

Fringe Tech 22 Jul 2026

GLM-4.7-Flash on One Consumer GPU: Why It Runs My Homelab Agent

A ~31B Mixture-of-Experts model at 4-bit, on one 32GB card, that beat a dense 32B on the real BFCL function-calling benchmark. What it is, what it is good for, and the numbers.

8 min read

Fringe Tech 22 Jul 2026

LLMs from the Ground Up: From One Number to a Fine-Tuned Model

Every core LLM idea, defined in order and built on the one before it: weights, tensors, attention, quantization, LoRA.

17 min read

Fringe Tech 20 Jul 2026

A Raspberry Pi Agent You Text Over Signal: The Complete Build

signal-cli from scratch on an ARM64 Pi, an agent runtime built from source for the Signal channel, and a read-only security model enforced in layers. Everything needed to rebuild it, including every trap.

10 min read

Fringe Tech 16 Jul 2026

The AIPI Lite, Rebuilt as a Fully Local Voice Assistant

A cheap cloud-locked AI gadget, reflashed to answer from your own LLM. No cloud, no subscription.

12 min read

Fringe Tech 09 Jul 2026

The whole library, offline, on one small computer

One small computer, a few watts, and a 290-source offline library that keeps working when the internet is off, and comes with you when you travel.

15 min read

Fringe Tech 04 Jul 2026

A Private Coding LLM I Rent Flat-Rate Through Proton Lumo.

I point my coding agent at Proton Lumo for a private model with no per-token bill, riding a subscription I already pay for. The provider block, the flat-rate plans, a forty-line proxy that proves which model and tier Proton served, and an honest read on where it works.

8 min read

Fringe Tech 04 Jul 2026

Build Your Own Weather Service: A Self-Hosted App That Stays Up When Its Upstreams Don't

Forecasts, radar, severe alerts, and a burn-window verdict from free public data, cached locally and reachable from anywhere, on hardware you already own.

15 min read

Fringe Tech 02 Jul 2026

A Private Coding LLM on My Own Network.

I run Qwen3-Coder on a spare RTX 5090 and point my coding agent at it over the LAN. Private, fast, no per-token bill, secured with an API key, and watched on a dashboard. Here is the whole build.

14 min read

Things I Love 22 Jun 2026

The Network Comes With You

A $79 Ubiquiti box the size of a deck of cards that makes a vacation rental behave like home, and why the unimpressive spec sheet is beside the point.

3 min read

Philosophy 22 Jun 2026

Trusting Your Gut Without Fooling Yourself

The first two posts said distrust yourself. This one says trust your gut. They are the same discipline. Kahneman, Klein, Simon, Ericsson, and the two questions that tell a real hunch from ego.

8 min read

Essay 22 Jun 2026

How to Change Your Mind Without Becoming a Weathervane

Two ways to fail: the weathervane that turns with every gust, and the bunker that never turns. Five moves from Darwin, Popper, Tetlock, Munger, and Feynman for being a compass instead.

9 min read

Philosophy 21 Jun 2026

I Reserve the Right to Change My Mind

Fillmore wrote the right to change his mind into a creed, Montaigne into a margin, Lincoln into a war. On principled revision, and the test that separates it from cowardice.

6 min read

Essay 21 Jun 2026

The colonoscopy study is right. The lesson is not the obvious one.

Expert endoscopists' unaided detection rate fell from 28.4 to 22.4 percent after a few months beside an AI. Here is one reading of why, and what it does and does not prove.

6 min read

Fringe Tech 21 Jun 2026

A Cistern Dashboard That Warns Me Before It Runs Dry

I run my place off an underground cistern, and a dry run can boil the pump and flood the basement. Here is the whole chain that put the tank on a dashboard, from the 4-20mA probe to the gallons reading on my phone.

17 min read

Fringe Tech 20 Jun 2026

My Home Dashboards, From Anywhere. No Ports Opened.

Reaching a LAN-only Grafana dashboard from anywhere, behind a login, without opening a single inbound port at home. A self-hosted WireGuard tunnel with Pangolin.

7 min read