AI engineer · AI agents · Automations.

RetroChainer retweeted
$6,000 IN. $24,789 OUT. FOUR DOTS RAN MY TRADING DESK, AND NOT ONE OF THEM CAN MOVE A DOLLAR WITHOUT ME. this is dotstrading. four dots, six strategies, one desk. us equities, fx, futures. six agents, six jobs: → atlas routes and fills → vega scans → orion reads sentiment → nova prices → titan holds the kill switch → luna settles the owner gate stays red until i open it. atlas never moves money alone. i kept the boring job for myself: saying yes. session closed at +$18,789. 13 fills, 0 pauses. one fill made about a third of that by itself. it's on screen in the clip.
1
26
840
RetroChainer retweeted
$100 A MONTH GETS YOU A DOT THAT RUNS ON ITS OWN CLOUD COMPUTER. CHAIN TEN TASKS AND 19.7% OF ITS RUNS GET FLAGGED. I CUT THE CHAIN IN TWO. openai launched dots on september 29 on gpt-6 astra. always on, 4,000+ apps, and the first one comes with Pro at $100, $200 or $500 a month. my field guide starts from two numbers: → five tasks in a row: 8.6% flagged → ten in a row: 19.7% twice the steps, more than twice the flags. so it's one dot, four seats, and a checkpoint every five steps. two legs of five come out at 3.4%. lookout, maker, skeptic, runner. the dot names its seat at the top of every reply. the skeptic closes each leg with five questions. the fifth one is "if i approve this and it is wrong, what breaks."
2
3
27
914
RetroChainer retweeted
$2,047 IN. $7,183 OUT. FOUR DOTS RAN MY TRADING DESK, AND NOT ONE OF THEM CAN MOVE A DOLLAR WITHOUT ME. this is dotstrading. four dots, six strategies, one desk. us equities, fx, futures. six agents, six jobs: → atlas routes and fills → vega scans → orion reads sentiment → nova prices → titan holds the kill switch → luna settles the owner gate stays red until i open it. atlas never moves money alone. i kept the boring job for myself: saying yes. session closed at +$5,136. 13 fills, 0 pauses. one fill made about a third of that by itself. it's on screen in the clip.
11
13
95
8,355
RetroChainer retweeted
$2,000 TO $20,000 A SHOT FOR THE WATER ALONE. THIS ONE HAS TWO VIKING SHIPS ON FIRE RAMMING EACH OTHER, AND A MAN FALLING OFF THE SIDE. two longships close in across open water, flames on both decks, a red flag on each mast. someone is already going over the side before the hulls touch. then the planks crunch together, spray goes up, and a whirlpool opens where he landed. my favourite detail is the sails. one stays untouched while the deck under it burns. a shot like this is one prompt on @Picsart now. no ships. no stunt team. no wet cameras. the last second is the quiet one.
9
3
68
3,540
RetroChainer retweeted
$5,000 IN. $15,677 OUT. FOUR DOTS RAN MY TRADING DESK, AND NOT ONE OF THEM CAN MOVE A DOLLAR WITHOUT ME. this is dotstrading. four dots, six strategies, one desk. us equities, fx, futures. six agents, six jobs: → atlas routes and fills → vega scans → orion reads sentiment → nova prices → titan holds the kill switch → luna settles the owner gate stays red until i open it. atlas never moves money alone. i kept the boring job for myself: saying yes. session closed at +$10,677. 13 fills, 0 pauses. one fill made almost half of that by itself. it's on screen in the clip.
6
16
76
10,992
RetroChainer retweeted
Tavus gave me early access to Griffin, so i did the most normal thing possible. i showed it my new hair. it caught the green right away. "okay okay, i see you. that's a major upgrade from just green. it sounds way more expensive." then it called it "a whole dual-toned aesthetic." when i said bye, it said "take care. good luck with the whole hike." i'd mentioned the hike earlier in the call. i actually said "oh, you remember that" out loud. not scripted. 51 seconds, recording below. tavus.io/griffin @tavus @hassaanraza
Introducing Griffin, the first model to pass the video Turing test. 48% of people who talked to it live thought it was a real human. Previous systems have had a pass rate <3%. It is #1 on NVIDIA's benchmark for full-duplex AI video. It’s the first Human Interaction Model (HIM).
Community note
The 48% figure and "video Turing test" claim are from Tavus's own study of 54 one-minute calls, not independently verified or using a standard protocol. Griffin-Lite leads NVIDIA's VideoFDB benchmark on their public leaderboard. cellcog.ai/blog/tavus-gri… research.nvidia.com/labs/amri/proj… tech-ish.com/2026/10/02/tav…
3
43
1,240
RetroChainer retweeted
a fat anon in a black hoodie just posted +$570 from his first test hour on fomo 10,000+ charts, data from 5,000+ wallets and over 3 weeks of development behind the mask. the terminal maps the proposed workflow: → charts - read price and volume → wallets - add context behind the activity → filter - separate a setup from noise → risk - define the limits before entry → trades - review entries and exits → record - preserve the result, including the ugly ones still early testing. three winning receipts look good on the wall. the next hundred trades matter more.
we provide - anon ai first test hour on fomo: +$570 you can see: fomo.family/profile/anon_ai still testing. the first wins are in. now comes the longer trading history. - trained on 10,000+ charts - data from 5,000+ wallets - over 3 weeks in development
1
1
25
387
RetroChainer retweeted
A GUY TIPPED A BUCKET OF CORN OUT OF A HELICOPTER ONTO LIVE LAVA. A PRIVATE CHARTER TO SHOOT THAT OVER HAWAII STARTS AT $4,132. the bucket is a plain grey plastic tub. the doors are off. the cracks under him are glowing orange. the kernels hit one crack and start popping on the spot. then a white column of popcorn starts climbing back up toward the door. this one is AI. a shot like this is one prompt on @Picsart now. no charter. no lava. nobody vacuuming the seats after. watch his face in the last two seconds.
6
3
38
3,657
RetroChainer retweeted
$3.02 A MONTH ON THE API → $0. MY JEVBRAIN NOW MAKES ITS CALLS ON MY MAC, AND IT JUST STOPPED CLAUDE FROM WRITING A SECRET INTO APP/KEYS.TS. this is jevbrain in no upload mode. claude builds. a local model decides. the model is von 1.0. 395M params, about 1.5 GB, around 18 ms a call, bound to 127.0.0.1. nothing goes out. before claude writes or edits a file, von scores one thing: is this a real secret? → 0.80 or higher: deny → 0.30 or higher: it asks me → lower: allow tonight: 362 allowed, 39 asked, 14 denied. the upload stat on its card is 00, and that's the one i care about. watch the popup at the top near the end. it says no, then tells claude what to do instead.
3
6
49
2,433
RetroChainer retweeted
A 15-SECOND MOTION AD RUNS $1,500 TO $3,000 AT A STUDIO. THIS HARNESS HAS OPUS 5.5 WRITE ONE AS CODE, THEN GRADE ITS OWN FRAMES LIKE A HARSH DIRECTOR. that clip is motion-studio, a harness built around opus 5.5 for one job: a 15s vertical reel with real UI in it. the model writes code, not pixels. → a headless browser walks the clock, 900 screenshots → ffmpeg stitches them into one mp4 → the soundtrack sets the timing, big moments land on downbeats two things are banned in its CLAUDE.md: a centered title on a gradient, and everything fading in together. then it grades its own frames: hook, legibility, motion, variety, sound sync, composition, brand. no full render until every score hits 8. round 1 had motion at 4. no timers. no css transitions. no Math.random. the last comment on the screen explains why it took three rounds.
5
4
40
2,072