amadeus

if you know, you don’t know

either the ai overlords are coming for me first or i'm getting cancelled for the shit i say to it

spent 3 years of my life working to this solution, and he just… tweeted it out.

@coral9000 ulf kragner bruk

niels jonker brug

@jlongster we had to add a second channel cause @fat always takes coworking

funny thing about using discord for team communication is you can always see whenever people are meeting radical level of transparency x.com/ryanvogel/stat…

new board I put together last night /cc @maxvoltar

@SlexAxton no, while i was watching the video of pewds train his new model

@amadeus Did you post this at dinner with me

failure is the meaning of life

@petersellis My wife, unfort

Have you ever met someone who was like, “yeah dude, I love Joe and the Juice.”

i don't think there's been any electronic/physical device/car/whatever the fuck in the last 20-30 years that is even comparable to this, aesthetically.

fax if tru

@thdxr 🫂

one place i hadn't been that diligent in curbing the chaos that LLMs create was in our tests looking through them gives me a preview of what happens when these things go unchecked and damn is it bad lot of ALL CAPS messages needed

do you admire the effort? orr...

OK — now I understand the full picture.

@thdxr i do that constantly... partially also because i feel like i'm really slow at understanding how code works, it's almost like i have to tinker with it to understand it

i think i still get the good feelings from programming even when using AI because i'm constantly refactoring i'll have it do a first pass and i'll look at it and think of a way to reduce it to less code that is more beautiful both fun and effective

they don't make 'em like they used to youtube.com/watch?v=4wgkP9…

@jlongster @thdxr you’re an all star

@thdxr hey now

pretty much every team in the ai space is smarter than ours but we're so much better looking it's outrageous

@thdxr what if i'm less smart and worse looking tho?

@petersellis @aakashgupta i appreciate your new study level military story arc peter

Narrator: it was not a dumb question.

The missile is reading the Earth like a blind person reading braille. Running its fingers across the surface and matching what it feels against what it memorized. Before GPS, engineers had a problem. You’ve got a cruise missile flying 500 mph at treetop level toward a target 1,500 miles away. No pilot. No satellite link. How does it know where it is? Their solution: give the missile a topographic map and let it read the ground. TERCOM (terrain contour matching) works like this. Before launch, you load the missile with a series of altitude profiles of the terrain along its flight path. Think of it as a cross-section of every hill, valley, and plateau between point A and point B, sliced into strips. During flight, a radar altimeter on the belly of the missile pings the ground constantly. It measures the exact altitude of the terrain below and builds a real-time profile. Then the onboard computer slides that real-time profile across the stored map, looking for the best match. When it finds the match, it knows exactly where it is, and corrects course. The wild part: this worked over enormous distances in the 1970s. The Tomahawk cruise missile could fly 1,000+ miles through a pre-programmed corridor of terrain strips, checking its position every few minutes, and arrive within meters of a target. No GPS. No external signal. Just ground texture. Then it got crazier. Engineers added DSMAC, digital scene-matching area correlation. Same concept, but with a camera instead of a radar altimeter. In the terminal phase, the missile takes a photograph of the ground below, digitizes it, and compares it to a stored satellite image of the target area. Pixel by pixel matching. In the 1980s. The engineering constraint that made all of this necessary is the interesting part. Inertial navigation systems drift over time. Gyroscopes accumulate tiny errors with each passing minute. Over a 2-hour flight, those tiny errors compound into hundreds of meters of deviation. TERCOM exists because engineers needed periodic “reality checks” to reset the drift. Every time the missile matches a terrain strip, it zeroes out the accumulated error and starts fresh. So the real architecture is: INS runs continuously as the baseline. TERCOM corrects the INS every few minutes by reading the ground. DSMAC does the final precision targeting by matching a camera image. Three totally independent systems layered on top of each other, each one compensating for the weakness of the others. GPS eventually simplified this stack, but militaries still keep TERCOM because GPS can be jammed. Terrain can’t be jammed. The ground is always there.​​​​​​​​​​​​​​​​ So they don’t “know.” They use the ground to calculate.

@cnakazawa i haven't opened vscode in years

Used up 70% of my (doubled) weekly codex limit in 24 hours. Have not opened VS Code. Shipped a lot, but every day is way more intense, too. My recommendation is to work intensely for 5-6 weeks at a time, then take a week off. Touch grass/snow, not tokens.

ready to smoke trees?