New AI releases, real tests, honest verdicts. I use this stuff daily and tell you what holds up.
Joined September 2026
- Tweets53
- Following58
- Followers23
- Likes28
AI Duck retweeted
This is f*king insane. Andrej Karpathy explained how every agent harness works, and he did it in a lecture about using ChatGPT
I cut his 2-hour "How I use LLMs" down to the 10 minutes that matter if you build agents
watch it next to the guide below and it clicks:
→ the context window is working memory. everything the agent knows right now is in there
→ the model itself is a lossy zip file. no calculator, no browser, no tools until you give them
→ every token costs twice: it slows the model down and it distracts it. keep it short, reset often
→ tool use is just a special token. the app pauses the model, runs the search or the code, pastes the result back
→ the model will quietly guess. his chart put 0.1 in for a missing year, then called a 20 trillion trend "1.7 trillion"
→ memory is a text file prepended to every chat. nothing magic
→ custom GPTs are just instructions + examples. "nothing special is happening here"
that's a harness. context, tools, verification, memory, instructions
most people fix a broken agent by changing the prompt or the model. this lecture shows why that rarely works: the problem lives in what goes in and out of the window
bookmark it and watch it tonight. then read the guide below and build your first one
This little tool turns pixel art sprites into normal maps
Drop in one sprite and it makes the normal map by itself. Your flat pixel art suddenly reacts to dynamic lighting like it actually has depth.
How it works:
• runs right in the browser, nothing to install
• works with any PNG you drag in
So you can just add it to the pixel art workflow you already have. No online service needed.
Small tool, but really useful if you want your 2D sprites to look deeper without repainting every light angle by hand.
AI Duck retweeted
A 23-YEAR-OLD CLAIMS 240 CLAUDE CODE AGENTS DID A $175K ANALYST'S JOB IN ONE AFTERNOON. THE F*CKING TRICK IS ONE LOOP
his post this week:
> 10 scouts, 9 extractors, 8 cross-checkers, 6 reconcilers, 3 judges
> 120 company filings
> every claim held against the exact source page
> anything that doesn't match goes back as "rewrite"
can't verify the kid. the loop is real and you can build it tonight
the rule underneath ↓ an agent never grades its own work
now point it at earnings season
your bot reads a press release, grabs "revenue up 12%" and trades it. one wrong number and you just bought a headline that never existed
the split:
→ the big model writes every claim with a page number
→ a cheap model opens that exact page and checks it
→ mismatch = rewrite, never a trade
→ only claims that survive both reach your signals
you only step in three times:
→ when a claim bounces twice
→ when the filing and the press release disagree
→ before anything goes near real money
the big model writes. the cheap one checks. nothing trades unverified
paste this into Claude Code ↓
"Build earnings_check.py:
1. Pull the latest 10-Q and earnings press release for [ticker] from SEC EDGAR.
2. Have Opus extract every number (revenue, EPS, margins, guidance) with the exact page it came from.
3. Have Haiku open that page and confirm each number. Mismatch → send back as rewrite.
4. Loop until every claim matches, max 3 passes. Log every bounce to verify.log.
5. Write only verified claims to signals.csv. Paper only, no orders.
Show me the code first. Don't run anything until I say go."
one of these every night. tomorrow's builds on this one
REA hit #1 on GitHub trending, and most people still run it inside a single chat
Split it into 4 subagents and one command takes a program you own from sealed binary to a tested rebuild with a match rate
Set it up once, aim it at the next one:
Holy sh*t, Opus 5.5 just made After Effects look ancient
No FL Studio. No Logic Pro. Not one sample
I gave it one idea and it made everything. The beat, the 3D scene and the ducks playing it
Every time a duck lands you hear it. Kick, clap, hats, bass, melody
This is f*cking insane…
Sound on and watch till the end ↓
CLAUDE IS INSANE AT DESIGN. BUT CHATGPT IS 7X CHEAPER
Asked Claude Opus 5.5 (left) and ChatGPT (right) for a launch page for a flying car.
ChatGPT: ~$0.03, done in 1:30. Claude: ~$0.20, done in 3 min.
ChatGPT gave me a clean dark template. Fine.
Claude gave me a 3D car that actually drives, lifts off and flies above the clouds. Plus a working altimeter and a paint studio that recolors it live.
7x the price. Way more than 7x the page.
Which one would you ship?
WHAT THE F*CK IS INSIDE THIS TESLA??
No jet engine.
No fuel tank.
No wings.
And it still flies.
So I X-rayed it 👇
- carbon fiber tanks in every empty space
-600 liters of compressed air
-10 thrusters that blast it out
- air goes down, car goes up
That’s it. It literally flies on the same thing you breathe
The only thing nobody knows yet: how long it can stay up there.
Would you ride it?
chatgpt doesn’t know what a car looks like
asked both to build a jelly g-wagon
-opus: real car that explodes into jelly pieces
-chatgpt: red bricks and 4 wheels
what exactly is chatgpt charging 2x for?
AI Duck retweeted
THIS IS STUPIDLY F*CKING GOLD
10,589 launches in 24h. 7,123 blocked. 52 trade signals. This Jev board has a trick almost nobody uses.
The idea: stop letting one model decide if a launch deserves an entry. Six agents score every launch, and the executor only fires when net EV after costs clears +3%. In the first 49 runs that's 18 entries, 27 skips, 4 blocks.
No gut calls, no FOMO entries, and every skip leaves a reason in the log.
You trace any decision back like this:
orb trace RUN-084199 --from-buy
here's the split:
→ Scout reads the feed at 41 events a second
→ Cartographer links wallets into 1,241 groups
→ Judge follows 32 wallets
→ Quant scores 87 setups a day
→ Guard blocks 72% of what reaches it
→ Executor keeps 1.8% in play against a 15% limit
You only look at three moments:
→ Guard blocks a launch because 45% of its buyers share one funder
→ Quant sees edge -10.6%, costs 3.9%, net EV -14.4% and skips
→ a launch passes Quant and Guard and goes to the executor
↳ animated concept with sample numbers, not a live P&L. the +3% after costs rule is the part worth copying
what the f*ck is with this duck
asked claude and chatgpt to build me a plush toy duck
claude gave him real fluffy fur, a felt bill, bead eyes and a squeak when you squeeze him
you can toss him, squish him, make him waddle. i’ve been playing with this duck for way too long
chatgpt’s version is right below 👇
cost: claude $6 | chatgpt $4.50
worth the extra $1.50?
kiwi jelly that actually feels alive — $4.50
not a video. not a render.
an interactive slice right in the browser.
glossy flesh, soft folds, real segments.
stretch it, drop it, watch it settle and bounce back.
one prompt in Claude.
I’ll run the exact same thing through every other model next.
this one already has me messing with it way too much.
Jelly King: Opus 5.5 Mogged Astra, Fable and Sonnet
In my jelly tests, Opus is in a league of its own right now. And it’s not just the wobble - the materials and detail are on another level too
Look at this citrus slice. Glossy flesh, visible segments, soft folds when you pull it. Astra bends too, but those sharp creases look more like a crumpled wrapper than jelly
Opus wins this one for me, and it’s cheaper. Maybe only Fable 5.5 can take the crown from Opus 👑
Estimated API cost:
Opus 5.5 - $7.57
ChatGPT-6 Astra - $10.15
Would you still pay more for Astra after watching this? 👀
real creators burn out. ai characters don't.
i broke down how jean phil went from zero to 275k followers - and the money behind it.
tools, prompts, real costs. no course, no paywall.
full tool stack inside - bookmark it 🦆
not every Peter Griffin makes an estimated $10k a week.
this one does - and he's not even real.
meet mr. paul grayson. no studio. no actor. no expensive production.
just a fictional face and a content machine.
here's the stack:
1. chatgpt images 2.5 - generate one face from multiple angles, keep it identical
2. give it a name, a backstory and a personality - not another generic ai model
3. take footage you have permission to edit
4. higgsfield - swap your character in, keep the movement
5. hook in the first 1-2 seconds and turn it into a series
people follow stories, not random clips.
talent or attention - which one wins in 2026?
save this before you build your first ai influencer 🦆
🤖 Made with AI
DEAD INTERNET THEORY IS REAL
this AI character's creator made an estimated $10K in just one week.
meet Mr. Paul Grayson:
→ 74.9K followers
→ 3.2M likes
→ roughly 26.1M views across 19 videos
→ 10.6M on his biggest clip
he looks like a real-life Peter Griffin - and he's trying to glow up.
awkward selfies, 'before' videos and attempts to look cool. a transformation storyline gives people a reason to follow the next episode.
here's how to test a similar format:
→ use ChatGPT Images 2.5 to generate your character from multiple angles
→ choose existing footage you have permission to edit
→ use Higgsfield or another character-swap tool to replace the person while keeping the movement
→ publish short episodes with a consistent personality
your first attempt might flop.
test several characters. build a network around the ones people watch and share, then explore sponsorships, affiliate offers and paid collaborations.
more experiments give you more chances - not guaranteed income.
bookmark this. build your first character and give people a reason to come back.
AI influencers are about to be huge.
People with no face are already clearing $70k+ weeks. And the craziest part? This is still early.
One character sheet.
Drop it into a video that already went viral.
Post it everywhere.
No camera. No studio.
The ones who start this month won't need to explain it in a year.
Who's actually building one?
there's people making more than $100k
with AI influencers
you can literally make stupid money so easily rn
that's the exact strategy:
- first you need to one shot
your ai influencer on justclip
- then you hop on any platform
that runs watermark campaigns
- then find viral trends and
swap your character on those
with justclip it handles the entire process
these AI characters will gain billions of views
the next couple of days
this is literally a golden opportunity
to make money online
$76K from a face that doesn't exist.
no camera. no studio. no voice. just a character sheet dropped into videos that were already going viral.
the AI character era just got its first receipts. or its first marketing.
which one do you think it is?
Сreators burn out. characters don't.
The next generation of millionaires won't have a face. they'll have a character.
A guy who doesn't exist has 209K followers. no camera. no studio. no bad days. just a phone, a laptop and a face people can't scroll past.
The playbook is now free. 919 bookmarks in under a day.
Most people will save it and never open it again. a few will build. those few are the new generation.
Which one are you?
