Talvion

[Talvion @Talvioon]

MIT audited $40B of corporate AI spending and found 95% of it earned nothing back.

The 5% that worked ran on budgets under $100,000.

The NANDA team went through 300 public deployments, sat with 52 organizations, and surveyed 153 executives. Every failed pilot had the same shape. Big vendor. Long procurement. A demo that impressed a boardroom in 40 minutes and never survived a Tuesday.

The gap wasn’t model quality. Every one of those companies had access to the same frontier models.

The gap was memory. The failed tools started every task from zero. They didn’t hold context, didn’t learn the workflow, didn’t adapt to how the team actually worked. Month 6 looked exactly like day 1.

Then the part that should scare a CTO.

Inside those same companies, employees were paying $20 a month out of their own pockets for ChatGPT and getting more done than the seven-figure internal build sitting two floors up. Nobody approved it. Nobody tracked it. It just worked.

Buying from an outside vendor succeeded roughly twice as often as building in-house. 67% against 33%. The instinct to own the stack cost more than it earned.

And over 50% of AI budgets went to sales and marketing, where the returns were thinnest. The money showed up in back office instead. Fewer agency retainers. Fewer outsourced contracts. Nothing you can put in a keynote.

So the question is which company you’re in right now: the one running a $2M pilot nobody opens, or the one where somebody’s personal $20 subscription is quietly doing the work.

The winning stack cost $20 a month. Nobody approved it.

[Talvion @Talvioon]

https://t.co/PpK6O6BI27

[Talvion @Talvioon]

@liamrecall By all means, feel free to save it.

[Talvion @Talvioon]

A professor in a sweater, 1 chalkboard, 0 animations and MIT’s calculus series out-pulls funded education channels.

4 minutes. No B-roll. No cutaways. Not a single frame from the actual lectures.

Gilbert Strang sits in a chair and talks. That is the whole production.

The linear algebra series went first. It moved so well that MIT greenlit calculus on the strength of it 5 videos for the big picture, 12 more on derivatives.

Education runs in the top CPM band. Adult audience, brand-safe, high intent. Advertisers pay for it.

MIT built the asset with a foundation grant and licensed it Creative Commons, non-commercial.

Which means the ad inventory sits with whoever re-uploads it, not with the people who filmed it.

Zero production budget, or zero rights to the upside pick which one you’d trade away.

He closes the video asking for donations.

[Talvion @Talvioon]

https://t.co/PpK6O6BI27

[Talvion @Talvioon]

@CryonixCrypto His lectures and lessons are truly incredible

[Talvion @Talvioon]

https://t.co/PpK6O6BI27

[Talvion @Talvioon]

A 100-SECOND FROG POSE CLIP FROM VIETNAM SELLS MORE MATS THAN A $50,000 CAMPAIGN.

The brand is named once, at second 92. It’s on screen for the other 97.

That gap is the business.

The clip looks like free advice. Warm up first. Hands perpendicular. Hips slow. Real technique for a real complaint everyone says the pose burns their thighs.

Nothing is sold for 91 seconds.

Then one line: use a mat, not knee pads. Purple, logo turned to the lens, sitting under her like furniture.

That’s 种草 (seeding). You don’t pitch. You fix one painful thing on camera, and the fix happens to carry a logo.

The math. Mat retails near $18. TikTok Shop affiliate takes 10-30%, call it $3 a unit. 800,000 views, 0.4% convert, 3,200 mats. Roughly $9,600 from a clip shot against a bedroom mirror.

Then it keeps selling for 8 more months. How-to doesn’t decay the way a livestream slot does.

So pick one: 6-hour livestreams with 70% to the agency, or 40 evergreen tutorials with the product on the floor.

The best ad in the video is the thing she’s kneeling on.

[Cryonix @CryonixCrypto]

https://t.co/vppYt2pilt

[Talvion @Talvioon]

@hitu_monke at first glance, the dog behaves very strangely, and human imagination draws strange things

[Talvion @Talvioon]

THIS DOG PULLED MILLIONS OF VIEWS PURELY OFF HUMAN IMAGINATION.

Not the footage. The gap in it.

Nicholas Epley found that we project a mind onto something under 3 conditions. It has to look a little like us. We have to want the connection. And we can’t know what’s actually happening inside it.

That third one carries the weight. Certainty kills projection. Ambiguity feeds it.

A human face on the same feed gives you words and context. Nothing to invent. A dog gives you a blank.

40 comments, 40 different inner monologues, all pinned on one animal that was probably just warm and bored.

You didn’t feel something for the dog. You wrote it, then read it back.

[Cryonix @CryonixCrypto]

https://t.co/vppYt2pilt

[Talvion @Talvioon]

A 67-SECOND KOREAN GYM CLIP OPENS WITH 5 SECONDS SHOT ENTIRELY FROM BEHIND.

Then the counting starts.

Hyperextensions on the bench. Light grey leggings. The camera does not move once.

Second 5. Smith machine squats. The trainer sits off to the side in a blue cap, MAYRK across the chest, calling depth.

Second 15. He strips plates and says he is going lighter. She stands in profile while he does it.

That is the pattern. Every working set gets a coaching line. Every rest gap gets a body shot.

Second 22. He demos the chest-supported row himself. An anatomy diagram cuts in lats, spinal erectors. Real information, 8 seconds of it.

Second 35. She runs the movement. He says squeeze your glutes. Yellow arrows drop on screen, pointing exactly where the lens already was.

Two audiences, 1 file. The ones learning hip hinge stay. The ones learning nothing stay longer.

Then the last 17 seconds arrive.

Korean text fills the frame. Roman chair, back extension, hamstrings and glutes. No subtitles. No handle. No price. No booking link.

The tank top says MAYRK. That is the entire brand in 67 seconds.

Matrix machines, clean light, vertical crop this thing is built to be saved and re-sent. It will clear 1M views on a slow week.

And it will sell 0 sessions.

Every frame points at her. Not 1 points at him.

[Cryonix @CryonixCrypto]

https://t.co/vppYt2pilt

[Talvion @Talvioon]

8 PEOPLE IN THE FRONT ROW CAUGHT SOMETHING ON THIS SERVE THAT ALMOST NOBODY WATCHING THE CLIP HAS NOTICED.

Court 2. 6:40 PM. No chair umpire, no Hawk-Eye, 30 folding seats half empty.

She tosses. Both feet leave the paint. The Wilson comes through clean.

Watch the crowd, not the player.

A man in a Christian Dior tee is out of his seat before the ball stops rolling, finger stabbing at the baseline. The blonde beside him opens her mouth and leaves it open. An older man in a blue cap waves them both off.

3 verdicts on 1 mark, and not one official in the building.

Go back and run it again. This time keep your eyes on the white line.

The front row was watching the ball the whole time.

Nobody else was.

[Cryonix @CryonixCrypto]

https://t.co/vppYt2pilt

[Talvion @Talvioon]

UNITREE'S HUMANOID BREAKDANCED FOR 19 SECONDS IN A ROOM FULL OF SUITS HOLDING PHONES.

No stage. No studio lights. Office carpet and a logo on the wall.

It drops low. Spins. Throws a wide lunge and holds it.

Arms circle. Legs open into a star. It goes almost flat to the floor, then comes back up on its own.

At second 11 it stands, takes 4 steps toward the camera, and walks past the crowd like a coworker heading to lunch.

Behind it: a quadruped parked by the wall. A red-and-gold mannequin. 2 screens playing empty roads.

The camera shakes the whole time. Frames get clipped. Nobody fixed it.

0 subtitles. 0 voiceover. No model name. No price. No call to action.

Now watch the footage again and look at the people instead of the robot.

Every one of them is holding a phone up.

Unitree didn't write that ad. The room did.

[Cryonix @CryonixCrypto]

https://t.co/0paRU9mGM2

[Talvion @Talvioon]

THIS 10-SECOND CLIP IS AI-GENERATED. THE FILE SAYS IT ISN’T.

Assembly floor. 6 female androids in a row, torsos open orange cable, green board, cyan light under clear plastic.

Technicians in grey stand back and watch.

A man in a suit reaches out. Presses the button on the first one’s chest.

Every LED in the row flips from cyan to magenta. 6 red cores ignite at once. Not hers. All of them.

She opens her eyes. Turns her head.

He kisses her. She kisses back. The other 5 keep facing forward.

Then the file itself: 576x1024, 30 fps, 303 frames, 2.3 MB, a “She Shorts AI” watermark burned into the bottom edge.

And one field buried in the container header. aigc_label_type. Value: 0.

0 means not AI-generated.

Every clip in your feed carries that same field. Almost nobody opens it.

The androids were transparent. The file wasn’t.

[Cryonix @CryonixCrypto]

https://t.co/0paRU9mGM2

[Talvion @Talvioon]

Tesla parked Optimus on a black podium and 40 phones went up in 6 seconds.

White gloss body. Black faceless head. TESLA badge on the chest.

A Cybertruck sits 5 meters behind it under showroom light.

A man in a straw hat narrates into his own camera. Nobody listens.

The shot is one take. 20 seconds, handheld, a slow circle around the podium.

Second 9, the camera reaches the back. Hip seams. Shoulder plates. The matte black skull.

Second 14, a man crosses his arms and just stares.

Then you notice it.

In 20 seconds of footage, Optimus does not move.

Not a step. Not a finger. Not a turn of the head.

No text on screen. No caption. No hands lifting anything. No walk.

The most-filmed robot in the room is standing on a turntable that isn't turning.

Every clip like this that goes viral shows the same thing: surface. Gloss, proportions, joint lines, badge. The parts that photograph well.

The parts that would prove it works stay off camera.

40 people shot a product demo where the product did nothing.

And it still traveled further than any lab video Tesla has released.

40 phones. 20 seconds. 0 movement. The future stood very still.

[BLAZT @BLAZT_Ai]

https://t.co/KyMr5S9Sfd