SOMEBODY LOOKED DOWN AT THEIR OWN BARE FEET AT THE TOP OF A TRANSPARENT WATER SLIDE THAT STARTED IN ORBIT AND ENDED IN THE PACIFIC
262,000 likes. 1,114 comments
The creator says in his own caption that it is fiction, generated, not footage of any mission or attraction. He says it before anyone asks
what the clip gets right, and almost all of it is the camera:
โ the feet. first-person POV with your own legs in the bottom of the frame is the single most persuasive framing available, because it is how you actually see
โ the curvature sits at the right distance. earth from that altitude has a specific horizon arc, and getting it wrong is the first thing anyone notices
โ the slide is transparent, which is a deliberate difficulty. it means the model has to keep refraction consistent against a moving background instead of hiding behind a solid surface
โ the lens flares behave like a GoPro, not like a film camera. wrong camera personality breaks the illusion faster than wrong physics
โ and it commits to one continuous fall. no cuts, because a cut in a POV shot is an admission
the POV frame is doing something specific here. it removes the actor entirely, so there is no face to get wrong, no performance to sustain, and no uncanny valley to cross. the only human element is a pair of legs, and legs are easy
which is why this framing is taking over. it is not that creators prefer first person. it is that first person deletes the hardest problem in generated video and replaces it with a problem about landscapes, and landscapes were solved first
the caption is the other half of the story. he labels it, in detail, voluntarily, and still clears a quarter of a million likes. the assumption that honesty costs reach keeps failing, and people keep repeating it anyway
if you want to feel where that line sits, image-to-video is the cheapest possible test: one still, one line about the motion. @Picsart runs it from a phone
no mission, no slide, no ocean. just the most convincing camera angle there is
AWS senior leaders take the stage across the week, sharing the launches and strategic direction that shape what organizations build next. The conversations those sessions spark become the architecture decisions you're making by the end of the week.
AARTI KIJE SHAIL SUTA KI,
JAGDAMBA KI AARTI KIJAI เฅฅ
S
N
E
H
S
U
D
H
A,
S
U
K
H
S
U
N
D
A
R
L
I
J
A
I
J
I
N
K
E
N
A
A
M
L
E
T
D
R
I
G
B
H
I
J
A
I,
A
I
S
I
V
A
H
M
A
T
A
V
A
S
U
D
H
A
K
I
AARTI KIJE SHAIL SUTA KI, @grok
JAGDAMBA KI AARTI KIJAI เฅฅ
๐จ GPT-6 ASTRA ANSWERED IN 18.2 SECONDS. I HAD TO FIX 92% OF THE RESULT. ๐จ
That was the pricing call.
A quick answer that handed almost the entire job back to me.
So I plotted all 10 tasks on two axes:
how long Astra took, and how much I had to fix.
The gap is brutal:
โ CLI utility: 9.8s / 6% fixed
โ docs site: 31.0s / 14% fixed
โ paper digest: 35.6s / 68% fixed
โ 3D asset kit: 48.2s / 72% fixed
The market scan took 41 seconds and needed 24% fixed. Iโd take that over the faster pricing answer.
Across this batch:
5 quick wins.
2 worth the wait.
1 fast mistake.
2 slow mistakes.
An agent can finish its turn while quietly starting yours.
That second axis shows how much work actually came back to the human.
Save this chart. Add a โhow much did I fix?โ column to your next AI test.