Misato
566 posts
Misato
@misat0x
writing about AI from somewhere beneath Tokyo-3
Tokyo-3 Katılım Haziran 2013
51 Takip Edilen114 Takipçiler
Jack Clark (co-founder of Anthropic) thinks Claude will start training itself by 2028.
"Claude 10 builds Claude 11. It designs the architecture, does the research, runs the training. We step back entirely."
what that means:
→ last 5-6 years of AI progress compressed into 2-3 years
→ then compressed again
→ humans out of the development loop
his 7-month-old will be in kindergarten when this happens.
bookmark this ↓
chuplung@choopyplug1
Anthropic added Claude to Slack. 65% of all PRs in their product org are now written by it. Boris Cherny (creator of Claude Code) has a Tag session running for a month. every day it checks data, fixes bugs, opens PRs. he just watches them come in. what makes it different from Claude Code: → you don't open it. it's already in the channel watching → multiplayer - whole team guides it, not just one person → remembers everything. tell it once, it never forgets → self-schedules work days or weeks out "I just got tired of tagging it. so I told it to always respond. now it just has my back." bookmark this ↓
English
Real-time face swap now runs on a single laptop at ~50ms inference.
Someone stacked face swap, pose detection, object recognition, and person profiling into one live pipeline, then piped the output straight into a virtual camera via UDP 00:02.
The whole system runs on consumer hardware and drops into any video call.
> real-time face swap at ~50ms inference
> pose + object detection running in parallel
> person profiling — age, clothing, mood, tattoos
> output streamed to a virtual camera via UDP
> runs on a single machine
The real story is not the swap quality. It is that this entire stack now fits on one laptop and works as a plug-and-play webcam.
Follow me so you don't miss out on trends in the world of AI.
kilo@kilo_cpa
English
built a machine that runs AI models locally and can pay for itself while you sleep
a serious AI user in 2026 is stacking Claude Code Max, ChatGPT Pro, Copilot, Cursor, Gemini Advanced
that's $450+ a month, $5,000+ a year, for compute you don't own and data that leaves your machine
running models on your own hardware flips that into a one time cost plus a few bucks of electricity
entry point is $249, a tiny NVIDIA dev kit that runs 7B models all day, covers most day to day tasks like drafting and quick coding
next tier is a $599 Mac Mini M4, its unified memory setup makes it punch way above its price on local inference
a used RTX 3090 for around $700 is still one of the best value picks out there, 24GB VRAM beats newer cards with less memory for this specific job
top tier is a mini PC AMD's own CEO personally signed at a developer event this year, it runs 235 billion parameter models locally in a box the size of a lunchbox
but here's the part most people building these rigs skip
the same GPUs can rent out their idle cycles when you're not using them
a high end 4 GPU rig running at 80% utilization can pull $700 to $1,400 a month on marketplaces like Vast,ai just from spare compute
AI companies are pouring over $600 billion into infrastructure this year and datacenter GPUs are still backordered for months
that's why idle consumer cards are suddenly worth something
one catch though, verification matters more than the hardware itself
an unverified card can sit right next to an identical verified one earning cents a day while the verified one pulls real money
so it's not just plug in and profit, uptime and reliability decide almost everything
would you rather run your own models on this or just rent the whole thing out?
Antid@antisadh
English
@misat0x lool are you also tired of what's happening in your feed?xD
English
PART 5. Building in public
I made my own MEDIA CONVERTER
What's up, guys?
I often have to edit videos and I'm really tired of searching for converter sites
And almost always, if I use a reference video, the audio track doesn't work
when I export it to DaVinci Resolve
I don't know why
But I have to extract it from various websites, including converting the video to another format after editing.
Of cource, they have free limits,
and you often hit them
So, I figured it wouldn't be that hard to create my own converter, so I decided to share it with you
It's pretty easy to install
If needed, I can create an installer to avoid any unnecessary hassle with GitHub
===================================
17 input types. 11 output options.
a small desktop converter with bundled FFmpeg
-> video: mp4, mkv, mov, webm, avi
-> image: png, jpg, webp, bmp, tiff
-> audio: extract WAV from video
No browser uploads
No separate FFmpeg install
share your projects in the comments!
===================================
github link:
github.com/svinosvin/Conv…
SSSvinosvin@SSSvinosvin
PART 4. Building in public Hey guys I spent some time working on the project again today Solo leveling on the minimum settings :0 Currently, we have 4 tabs 1) table In the table, we have a mini calendar that displays our tasks, sorted by week 2) tasks The tasks tab is where we can manage our tasks It still needs some work, and quite a bit 3) skills In the skills tab, we've added skills that we'll be upgrading in the future 4) lvls In this tab, we'll see our progress bar and skill level. ====================================== The most important changes: -> I added the ability to add the same task for multiple days so I don't have to do it manually. I had to spend a bit of time on it. -> And I added copy-and-paste functionality so I don't have to rewrite everything -> I also changed the app's color scheme, making it more visually appealing. I haven't decided on the color scheme yet -> Everything is already saved to the database -> The widget hangs as a subtask and is not displayed in the taskbar --------------------------------------------------------- Stack: Tauri (React + Rust + Tailwind.css) I write some things by hand, some things are helped by claude code, but now I try to do it myself I'm trying to get back to my roots and try to stretch my brain Thank you for reading! I appreciate your support! github link: github.com/svinosvin/todo…
English
The best World Cup position for me is Brazil vs Norway
I'm taking over 2.5 and here's why
Brazil are the favorite, but both teams have looked pretty good so far
and for totals, I'm mostly looking at the stats from their previous matches
Brazil
June 13: Brazil 1-1 Morocco
June 19: Brazil 3-0 Haiti
June 24: Brazil 3-0 Scotland
June 29: Brazil 2-1 Japan
total: 4 matches, 9 scored, 2 conceded
over 2.5 hit in 3 of 4 matches
Norway
June 16: Norway 4-1 Iraq
June 22: Norway 3-2 Senegal
June 26: France 4-1 Norway
June 30: Norway 2-1 Côte d’Ivoire
total: 4 matches, 10 scored, 8 conceded
over 2.5 hit in 4 of 4 matches
Norway conceded in every single match
the winner goes to the quarterfinals
the loser is out of the tournament
both teams have to fight, and I think we get a beautiful game here
btw, I decided to take this position after watching the new @DG3_terminal brand film
and of course, I'm taking it on DG3 too
I think it ends 2-1 Brazil
what score do you see?
DG3@DG3_terminal
man negotiates his own release after Argentina finally survives Vozinha's reflexes 👀 Ransom arrangement is now super easy. Predict the FIFA WC, and enjoy faster execution and instant withdrawals on DG3.
English
FABLE 5 TURNS AI FROM A CHATBOX INTO A WORKER YOU CAN LEAVE ALONE
Most people still use AI one message at a time.
They ask for a function, paste the error, ask for a fix, paste the next error, then ask for tests.
Twenty messages later the project works only because they acted as the motor.
Fable 5 changes that.
You give it the full outcome, the files, the context, the success criteria and a limit.
Then it plans, writes, tests, fixes its own errors and keeps going until the job is done or it gets truly stuck.
That sounds small until you realize what changed.
You are not prompting the next step anymore.
You are handing off the whole project.
This is why tiny tasks are the wrong use case.
Don’t use Fable 5 to rewrite an email, fix one button or explain one error.
Use it for the annoying work that normally eats an afternoon:
> build a small tool
> clean up a messy folder
> research 15 sources
> fix a codebase with tests
> turn a brief into a finished project
The new skill is not writing clever prompts.
It is defining what “finished” means and letting the model run.
Most people will keep chatting with AI.
Others will start delegating entire projects.
save this
Insomnia@insomnia_vip
English
I analyzed 100+ AI creator accounts. Here's what surprised me.
The biggest winners aren't building better prompts.
They're building AI personalities.
After looking at dozens of AI-first creator businesses, one pattern kept showing up:
→ 72,000+ followers built with short-form content
→ 510 paying subscribers
→ $8.99/month subscription
→ $4,580+ in monthly recurring revenue
The strategy is surprisingly simple.
Create an AI character people remember.
Post consistently on TikTok, Instagram, or X.
Turn attention into recurring subscriptions.
Most people think AI is about productivity.
The smartest creators use it for something much bigger:
Creating digital assets that attract an audience 24/7.
In the next few years, we'll see thousands of AI creators earning more than traditional influencers—with no camera, no office, and no team.
The question is no longer:
"Will AI replace creators?"
It's:
How long until your favorite creator isn't even a real person?
Bober_smart@Bober_smart
English
AN IDEAL EVENING WITH FRIENDS EXACTLY WHAT WAS NEEDED TO RECHARGE.
i always love weekends for this: the chance to relax with my closest circle, put all my worries aside, and just enjoy the moment.
we stopped for pizza, and of course, i couldn't pass up my favorite kronenbourg blanc.
then we took a long walk along the embankment.
the most unexpected surprise was meeting this little guy on the way home he didn’t leave me a chance to pass by, i had to stop to take a photo and pet this prickly miracle. 😊
and what about you? how did you spend your weekend, or what are your plans?
share in the comments!
Crypto Mavka@CryptoMavka
my brain officially took a day off 🌊 day 2 of relaxing sand, water and complete silence yesterday it was shelby, pizza, champagne and the girls today it's just me and the water turns out weekends exist for a reason who's also near the water today drop your photos or just tell me where you are right now 👇
English
I always assumed blockchain infrastructure was built mainly for blockchain engineers
Reading about @CNPYNetwork made me rethink that
With AI becoming part of how developers build software, maybe the next challenge isn’t teaching more people blockchain
Maybe it’s making blockchain easier for more builders to create with
Could the next generation of blockchain builders come from AI first developers instead of traditional Web3 teams?
English
"AI PREDICTED VOZINHA'S RISE
$1,700/month reading patterns others miss
I called it after Spain. Then Argentina happened."
"After Vozinha's performance against Spain, I asked Claude one question:
'Based on this data — what happens to this goalkeeper next?'
The answer came back in seconds:
— 20M+ new followers within 30 days
— Top 3 most searched athletes at World Cup 2026
— Probability of knockout stage heroics: HIGH
— Reason: underdog narrative + elite performance = viral inevitability
I posted it. Nobody believed me.
Then Argentina happened.
3-2. Extra time. Own goal in the dying seconds.
0.45 xG for Cape Verde vs 2.16 for Argentina.
Vozinha — 40 years old — frustrated Messi for 120 minutes.
19 years grinding in Moldova, Angola, Cyprus, Slovakia.
Sidny Cabral scored what may be the greatest goal in Cape Verdean history to make it 2-2.
An own goal ended the dream. Not Messi. A deflection.
That's how close a nation of 529,000 people came to eliminating the World Champions.
Claude didn't predict the scoreline.
It predicted the story.
Because AI doesn't watch football — it reads patterns.
And the pattern was always there:
7 clean sheets in 10 qualifying games.
500K followers before Spain.
17 million after.
The data knew before the world did.
Here's what people don't realize:
This is a skill you can monetize.
Sports analysis. Trend prediction. Viral moment forecasting.
I generate $1,700+/month doing exactly this — running Claude on public data, spotting patterns before they become headlines, and selling that insight.
No studio. No team. No expensive tools.
Just Claude, data, and the ability to ask the right questions.
Vozinha spent 19 years showing up when nobody watched.
I spent 3 months learning to use AI properly.
Both of us got results nobody expected.
The blueprint works.
Build the skill. Read the patterns. Get paid for seeing what others miss."
World Cup 2026@ofootball__
Thank You VOZINHA 🫡
English
Jordan Belfort, the "Wolf of Wall Street" who did time for fraud, now teaches a boring, safe formula instead.
A $1 billion hedge fund collects $20 million a year before it even starts, plus 20% of profits. And it almost never beats the S&P 500.
"Index fund on the S&P 500... some bonds, depending on your age... cash for emergencies... and let's say 5% for speculation," Belfort says.
"There's nothing wrong with speculating, it's fun. Just don't secure your retirement with it," he adds.
See below ↓
Riley West@rileywestreel
As a student, Carl Icahn lost a full week's pay in a poker game to the owner of the beach club where he worked. Two weeks later, after reading three books on poker strategy, he was winning $500 a week and ended the summer with $2,000, against the $750 he needed for room and board at Princeton. "The real or liquidating value of many American companies has increased markedly in the last few years; however, this has not at all been reflected in the market value of their common stocks. Sizable profits can be earned by taking large positions in undervalued stocks." From 1968 to 2011, he compounded an initial $100,000 at a 31% annual rate. Over the same period, Buffett's Berkshire Hathaway grew at 20% a year. "The consensus thinking is generally wrong. If you go with a trend, the momentum always falls apart on you. So I buy companies that are not glamorous and usually out of favor." In October 2012, Netflix shares had crashed 80% to $58, and Icahn put in $321 million for nearly 10% of the company. By his own account, the position was up 457% in 14 months, and over three years Netflix earned him close to $2 billion. See below ↓
English
AI didn't replace web agencies.
it replaced the hardest part of starting one.
Google Maps finds the leads.
Claude writes the copy.
AI builds the website.
you send the finished version.
suddenly...
a $1,000 client doesn't feel that far away.
Raytar@Raytar
English
Bezos just raised $12B for an AI startup called Prometheus.
The pitch: compress 10-year engineering cycles down to 1.
That’s a bold claim. But the actual bet is simpler and more interesting.
Most AI went after software. Prometheus is training on physics, simulations, and manufacturing data.
That’s a different problem with a different ceiling.
The “AGE” framing is marketing. The underlying direction is real.
What industry do you think flips first when design cycles compress by 5x?
English
I've been trying out the new @Jggl_it iOS app, and it's more than just another social platform.
What makes it interesting is the AI integration.
Instead of only scrolling through content, you can interact with AI to generate images, videos, and music.
You can also build your own profile, connect with other users, and explore features that combine social networking with AI-powered creativity.
I just finished setting up my profile and I'm looking forward to seeing how the ecosystem grows from here.
My JGGL profile
jggl.ai/?user_id=01KW2…
#APP_JGGLIsHere
English
As a ghostwriter, AI is one of my favorite tools to use.
But its getting more complex everyday and its hard to see whats good for you and not.
especially for web3 research, its getting more important too.
lets see what we got this week 👇
@claudeai shipped a lot this past week and most people know half of it.
- two new models landed. sonnet 5 (June 30) is the most agentic sonnet yet, planning and running tasks on its own that needed pricier models months ago.
- the big one is back. Fable 5 (July 1) is Anthropic's most capable public model, beating Opus on long, complex tasks.
- don't miss this: through july 7 you can spend up to 50% of your weekly limit on Fable 5. Give it your hardest task this week.
- skills are the cheat code. teach claude your format or brand once and it triggers itself, so you stop re-explaining your process every chat.
- weird trick that works: for big tasks, tell Claude to interview you first. The output beats jumping straight in.
- set it and forget it. In cowork, tell claude once to run your daily email check or weekly report, and it handles the rest.
the model got smarter, but the edge is setup.
People winning with claude aren't prompting harder, they're building systems that run without them.
which one are you trying first?
English
A guy I know bills $12,000 a month writing prompts for companies. His whole method is one free Anthropic session.
Two Anthropic engineers on stage, walking through a real customer case. Prompting 101.
The opening is almost embarrassing. Claude reads a car crash report and decides it's a skiing accident.
By the end of the video, the same Claude reads the same report and delivers verdicts an insurance company can ship to production. Everything that changes in between is the method.
It's a 10-part prompt structure Anthropic uses internally. They build it live, piece by piece, and show what each piece fixes. The part about ordering, near the middle, is what my guy called "the whole game."
He watched it twice. Second time with the console open.
Most people write prompts like messages. Anthropic writes them like contracts.
The gap between those two is $12,000 a month.
Moysei@0xMoysei
English