synapto retweeted
remember when everyone was parking money in anchor on terra for like 20% a year on a stablecoin
sounded safe right its a dollar coin it doesnt move
over 14 billion dollars sat in there
what nobody really read
that 20% was supposed to come from people borrowing but barely anyone was borrowing compared to how many people were depositing
so they just kept topping up a reserve to cover the gap
70 million once, then 450 million more
in may 2022 the money ran out and ust and luna went down together
if someone has to keep pouring cash in to pay your interest, you're not earning yield, you're getting paid to stay until they stop
so now whenever i see a juicy apy i just ask
who is actually paying me this and what happens when they stop
@RallyOnChain this is the fine print nobody reads
I wrote one prompt: "top-down cartoon ducks playing basketball, buzzer beater, slow-mo, confetti
AI turned it into 30 seconds at 60 fps. It drew every frame in code and synthesized all the sound, bonks and swishes included
The rim roll at 0:24 is my favorite part
What would you prompt next?
2
6
18
synapto retweeted
2010: we consumed content
2020: we created it
2026: we generate it with ai
2035: ai runs the business
which step are you on right now?
synapto retweeted
Zoom into "I'm not a robot". It never ends.
That checkbox exists to catch machines. Machines now pass it better than we do: in a 2023 UC Irvine study with 1,400 people, humans solved distorted-text CAPTCHAs 50 to 84% of the time. Bots got almost 100%, in under a second.
So the video follows a machine trying to look human: captcha, server rack, a chip that looks like a city, a laptop at night, 99% buffering, 1% battery, Snake, a 3D hand with one finger too many. Then back to the fresco.
From the first frame to the loop you zoom in about 500 billion times. The fresco is the real Michelangelo, every other scene is drawn in code.
synapto retweeted
Everyone says Claude can replace an assistant. So I gave it 15 real work emails and one prompt.
It caught a cancelled request buried in a follow-up, found a deadline hidden at the end of a long email, and converted time zones on its own.
It also dropped a late fee and ranked a contract that dies Wednesday below the morning standup.
Fast first pass. Not a replacement.
synapto retweeted
i asked claude to build a robot panther and take it apart piece by piece.
eyes, claws, spine, battery, every joint. then it walks and pounces out of frame.
no blender. no video generator. just code 👇
synapto retweeted
This is one workflow, built by hand, node by node.Somebody sat there and wired every connection: this output goes into that input, this branch handles the error, this one retries. Hours of work, and it only does one job. Change the job and you are back in there moving wires.Agents are eating this part. Not the thinking, the wiring. You describe what should happen and the plan assembles itself. Which means the valuable skill moves from knowing the tool to knowing what you actually want.How much of your week goes into wiring instead of deciding?
OpenAI let AI agents loose on the Linux kernel for 10 days
They came back with 948 working root exploits.
Here's what actually happened
The setup
an unreleased OpenAI cyber model
2.8M lines of kernel code (about 7% of it)
agents found the bugs, tested them and wrote the exploits
The funnel
2,308 candidate bugs in 12 hours
1,724 reproduced in 30 hours
948 distinct privilege-escalation flaws with working exploits
The price
$1.21M in API credits. About $1,280 per exploit.
One Linux root exploit pays $20K–$70K at hacking competitions.
In July their previous run found 166 for ~$3,100 each.
Three months later: 5.7x more, at less than half the price per bug.
Now the part most people will skip:
1. Finding bugs is no longer the bottleneck. OpenAI says the hard part was figuring out which findings were real.
2. Agent "wins" still need checking. An early run claimed 200 root exploits. 20 of them actually got root through a different bug than the one assigned.
3. Fixing is way slower than finding. After 10 days: 61 patched upstream, 43 in staging. Out of 948.
Fair caveat: every exploit needs an existing foothold on the machine. These aren't remote hacks.
But OpenAI's own line says it all: a low-privileged account on an outdated OS "should be considered functionally equivalent to root access."
My takeaway: finding bugs just got cheap. Verifying and patching them is the new moat.
And if your coding agent has shell access on your laptop, sandbox it.
Who wins this race, attackers or defenders?
synapto retweeted
Green screens are officially dead.
Kling 3.0 just turned an empty white wall and a black
t-shirt into a full cinematic detective thriller-while keeping every subtle head movement and sip perfectly synced.
No set design. No lighting rig. No wardrobe budget. Just raw footage and AI restyling.
Indie filmmaking will never be the same.