Duke Ian

@romevsjude @MorgothsReview Do you even know what a bear trap is? It is a set of metal jaws, designed to bite down and trap a bear, so the bear either dies or chews off its own leg to escape.

@MorgothsReview 🫡

Donald Trump’s meltdown after marching his forces into the most glaringly obvious bear trap in the history of human conflict is the most undignified, embarrassing episode I’ve ever seen in politics.

@magills_ @TexasCalvinball @akevittnvg @TJtheSting gas station fried chicken is also amazing, in the south.

@TexasCalvinball @akevittnvg @TJtheSting Our gas stations serve breakfast pizzas

The X timeline right now

I'm not vibe coding, I'm vibe texting

President Trump Participates in a Bilateral Meeting, Mar. 19, 2026 twitter.com/i/broadcasts/1…

# Slide 4: Analog — Continuous Signal + Noise Model ## Style Brief (Same as slide 1 — warm parchment, pencil sketch, dense lab notebook feel.) ## Layout: Encode/Decode Split with Margin Notes ## Prompt Technical pencil sketch on warm aged parchment paper with subtle coffee stain marks, worn edges, and faint ruled lines showing through like notebook paper. Dense hand-drawn diagram in dark brown ink with slightly uneven hand-lettered labels. 16:9 aspect ratio. The composition should feel like a dense page from a researcher's lab notebook. TOP: Bold hand-lettered title "4. ANALOG — CONTINUOUS SIGNAL + NOISE MODEL" centered, with a double underline. Smaller subtitle: "signal grows linearly, noise grows as square root of N". The main area is divided into upper ENCODE and lower DECODE sections by a hand-drawn horizontal dashed line. UPPER SECTION labeled "ENCODING" in a sketched tab: Far left: A pinned note card titled "PER-FRAME FORMULA" containing a large hand-lettered equation: "frame[cell] = sign * (S + payload_bias) + noise" with annotations below each term: "sign" has arrow to "+1 or -1 from QR", "S" has arrow to "signal_strength", "payload_bias" has arrow to "+delta or -delta from payload bit", "noise" has arrow to "uniform random from PRNG". The card has a pushpin and slightly curled corner. Center: A large hand-drawn graph taking up significant space, titled "SNR IMPROVEMENT" in bold lettering. The x-axis is labeled "N frames" with tick marks at 1, 4, 16, 64, 256. The y-axis is labeled "accumulated value". Two hand-drawn curves: a steep straight line rising steeply labeled "Signal = N * S (linear)" drawn in bold strokes, and a much flatter curve labeled "Noise = sqrt(N) * amplitude" drawn in thinner strokes. The growing gap between them is cross-hatched and labeled "SNR = sqrt(N)". Below the graph, a small table shows: "N=1: SNR=1" then "N=4: SNR=2" then "N=16: SNR=4" then "N=64: SNR=8". Right side: A hand-drawn 3D perspective sketch of a height field or terrain map, showing peaks and valleys arranged in a QR-like pattern. Tall peaks are labeled "+N*S (white module)" and deep valleys labeled "-N*S (black module)". A horizontal plane cuts through at zero labeled "threshold plane: sign = QR". The varying heights of peaks above the threshold are annotated "taller peak = payload bit 1" and "shorter peak = payload bit 0". Title above: "ACCUMULATED HEIGHT FIELD". Top margin annotation: "Grid, min 2 frames, signal_strength must exceed noise_amplitude" LOWER SECTION labeled "DECODING" in a sketched tab: Left side: A stack of overlapping frame sketches labeled "N float frames" with wavy lines suggesting continuous values, flowing into a large arrow labeled "sum all frames" leading to an accumulated grid with float values like "+4.8 -3.2 +5.1 -4.7". Center: A critical step shown as a large diagram. The accumulated grid sits at left. A second grid labeled "expected noise sum" sits below it, drawn with dashed borders and annotation "reconstructed from shared PRNG seed — same key, same noise". A large minus sign between them leads to a third grid labeled "CLEANED FIELD" with cleaner values. Annotation: "noise is deterministic, so we subtract it exactly". This is shown as the key insight with double-underline on "subtract it exactly". Center-right: From the cleaned field, two forking arrows. Upper arrow labeled "SIGN (threshold at 0)" leads to a QR grid with label "recovered QR". Lower arrow labeled "MAGNITUDE (distance from baseline)" leads to a comparison diagram showing actual magnitude versus expected baseline "N * signal_strength", with residuals marked as bit=1 (above) and bit=0 (below). A small majority-vote tally for multi-cell positions. Right margin: A small sketched note card titled "DETERMINISTIC NOISE" containing: "encoder seeds: Prng::from_key(qr_key, frame_index)" and "decoder reconstructs identical sequence" and "therefore: noise cancels PERFECTLY" with "perfectly" underlined. BOTTOM LEFT: Bold hand-lettered callout in a rough box: "FROM DISCRETE TO CONTINUOUS" with smaller text: "binary flipped biased coins. analog adds real-valued signal plus noise. now we have a formal noise model: SNR improves as sqrt(N). and deterministic noise means perfect cancellation on decode." BOTTOM CENTER: Properties table in a sketched frame. Left column "HAS": continuous float values, formal SNR model, deterministic noise cancellation, payload in magnitude, grayscale appearance. Right column "LACKS": multi-layer structure, overlapping windows, spatial permutation, boundary-free carrier. BOTTOM RIGHT: Torn-edge paper with arrow pointing right: "NEXT: what if L1 accumulated outputs THEMSELVES became frames for a second layer? Recursive steganography — a QR hidden inside a QR..." with "recursive" double-underlined. No people, no faces, no color beyond dark brown ink on warm parchment. Dense margin annotations, thin leader lines, pushpins on note cards.

@ianzepp @gauntletai Astonishing, frankly! Love to see the metaprompt and generated prompt for these, if you’d be willing.

Using Claude to generate prompts for Gemini & Nano Banana produces exceptional results. This is for a QR-focused temporal stenography side project using skills I've picked up in the @gauntletai program.

I am becoming more and more convinced that Typescript will shortly suffer the Sports Illustrated curse. Top of the game atm, destined to go downhill from here now that Rust is so easy to write with AI.

@TheSpiciestDev @gauntletai Leaving all the porsches at home. Bringing the kitted up Bronco.

@ianzepp @gauntletai Driving a fun car? 😎

Completed week 3 of @gauntletai and driving to Austin (17 hrs away) tomorrow. Project this week: chunking a COBOL project into a vector DB and hitting it with LLM queries, with a sub-3s performance requirement (I hit 1.5s consistently). Demo below.

@nakoerschner @gauntletai uh, no, i drink too.

@ianzepp @gauntletai So you'll be the DD while we're in Austin?

I haven't browsed rennlist in months and I go back now, and it's the exact same ego and ADM and "review my build" and boomer talk. But with chickens?

Hook: On top the regular project, I decided to rewrite the bulk of Ghostfolio (the wealth mgmt tool) in Go & Svelte, in under 48 hours. @gauntletai .. Here's how it went (hint: it went well). Crazy Eddie starts at 4:34

@erenbali Nailed it.

Every technical founder who had stopped coding 10 years ago

This looks like pixels. It is not. It is 250,000 individual canvas2d shapes streamed to my week 1 @gauntletai project UI. It is completely unusable, but so fun to get working.

One more: Dinosaur import / export!

Finally, and this is the last one: minimap HUD!

alright, so this is just going to be a flow of supposed consciousness, related to @gauntletai cohort 4 week 1. approaching the early-submit deadline (tonight actually), and rather than talk my way through way I've done, I figured I'd just dump a bunch of vids and screenshots. 🧵