Sabitlenmiş Tweet
PROFESSOR SNAPE HAS SEEN ENOUGH.
Kimi K3 vs GPT-5.6 Sol.
Same prompt. Two completely different spells.
Who won this round?
x.com/asymetx/status…
Asymet@asymetx
English
Asymet
720 posts
Asymet
@asymetx
Reading probabilities | Hunting narratives
Onchain Katılım Haziran 2026
186 Takip Edilen78 Takipçiler
IF YOU’VE BEEN USING OBSIDIAN… YOU’VE BEEN DOING EVERYTHING WRONG.
Does your second brain look like this?
A living neurological system with real neurons firing in real time.
Prefrontal Cortex. Motor Cortex. Concept Layer. Hippocampus. Sensory Cortex. All actively working and connecting.
This isn’t just fancy visuals — it has STDP learning, powerful pattern recognition, and actually thinks with you.
Not only that — your agents are directly connected to this brain. They do the tasks they’re trained for, feed new information back into the system, pull insights from it, and the cycle continues.
The whole thing just keeps getting smarter.
This is what a real, active second brain looks like in 2026. No more static notes. No more dead storage.
The full workflow is right below this post.
Semichenko@semichenkko
English
ONE FINGER TAP FREEZES A PERSON MID-MOTION. EVERYTHING AROUND THEM KEEPS MOVING
Not the whole world. Just the target - locked in place, real physics holding, while the frame around them stays completely alive
Claude structured the sequence first - the exact moment the tap lands, what freezes, what keeps moving, how long the hold lasts - before Seedance 2.0 rendered it frame by frame.
Under 10 minutes of render time. A 12 dollars in tokens
A localized time-freeze effect like this would normally need a VFX compositing team and a multi-week post-production pass.
No compositing team. No post-production pass. No green screen.
What would you freeze first?
Andy🤖@andy_neon_
English
STOP MANUALLY BABYSITTING YOUR CODING AGENT'S CONTEXT WINDOW. THIS GUY AUTOMATED THE ENTIRE THING AND IT NEVER STOPS RUNNING.
The core of it: a PTY (pseudo terminal), which lets you interact with a terminal programmatically instead of manually typing prompts.
Here's what that actually unlocks:
→ Instead of typing into Claude Code CLI, a script sends the prompt automatically
→ Works with any coding agent CLI – Claude Code, Open Code, Codex, Hermes – through the same programmatic interaction
→ Full scripted control over restarts, new sessions, clearing context, resuming old sessions – no manual slash commands needed
The clever part: you can monitor context consumption in real time and automatically hand off to a new session before it runs out – basically building your own version of auto-compaction, but with full control over exactly when and how it happens.
Then he wraps all these PTY terminals under a daemon running 24/7 – so every agent stays alive continuously, no matter what.
And the final layer: injecting prompts directly into these terminals, so scheduled tasks or external messages from Telegram or Slack can trigger the agent without him touching anything.
This is what "always-on AI agent" actually means at the infrastructure level – not a buzzword, an actual system.
Bookmark this. Full breakdown in the video below.
SCOTTY BEAM@ScottyBeamIO
English
GPT 5.6 bros went quiet, Opus 5 on GOD MODE is straight up CRIMINAL.
The exact contract to run it on, top to bottom:
1. pick claude-opus-5, effort High. thinking is on by default now, the model decides how deep to go, you just set the gear
2. write the finish line, not the route: "here's the outcome, here's what must never break, here's how we know it's done". zero step lists, its judgment beats yours this generation
3. pin five laws in one file: no special cases, boring stack, show me before anything irreversible, every "done" ships with proof, laws beat my instructions
4. now delete every "verify your work" line from your old prompts. it checks itself unprompted, those lines just make it check twice and bill you twice
5. the grader lives inside now: "split into a writer and a fresh-context verifier that assumes the build is broken, loop until the verifier goes quiet"
6. dump the whole archive in, the 1M window swallows your entire repo plus every old session log. it never starts from zero again
7. kick it off before bed with one rule: don't stop to ask, log every decision, leave the summary on top. the morning version of the project is just further along
The deeper version of every step (the full contract, the effort ladder, what to stop prompting, when max is worth the bill) is in the article below.
P.S. hand the article to your Opus 5 and say "this is our contract now".
Miraqle@0xMiraqle
Total breakdown on how to take your OPUS 5 to a GOD-MODE level that works as an army of teammates x.com/i/article/2080…
English
A 27-year-old man put 6 AI agents inside a fake space station and watched an Etsy store crawl toward $2,000 from one command chair.
The strange part is that the station is not just decoration.
6 agents. 1 dashboard. $2,000 target. 2 factories. 1 commander.
Each room shows a job: Ultron watches the whole operation, Nova sits in the research lab, and the dashboard tracks revenue, weekly performance, and what every agent is doing.
The Etsy store is still the main engine.
Nova keeps scanning products, trends, and business signals while the commander sees the whole system without opening 12 tabs.
That is the real idea hiding under the sci-fi skin.
Agents become easier to manage when the work has rooms, roles, and a visible control panel.
Thariq@trq212
We removed ~80% of the Claude Code system prompt for our newest models, this is what we've learned about writing system prompts, skills and Claude.MDs for them. x.com/i/article/2080…
English
Blender used to punish you for trying to make a 10-second cinematic. Now Kimi K3 can do most of the punishment for you.
A scene like this needs detailed lighting, reflections, camera movement, animation, and enough manual fixes to ruin your afternoon.
With Kimi K3 connected through Blender MCP, you can describe the idea in plain English and let it build the scene, write the Python, position the objects, animate the camera, and render inside the actual Blender project.
Then it can inspect the result, fix broken scale, weak lighting, or awkward movement, and keep iterating instead of starting over.
You still need taste. But the gap between “I have an idea” and “I have an editable 3D commercial” just got embarrassingly small.
Rina@irinatoxi
English
CHINESE DEV JUST STUFFED GPUs INTO A DEAD MINING RIG AND MADE $340K LAST YEAR WHILE HIS EX-COFOUNDERS BURNED SERIES A ON ANTHROPIC BILLS
Dual black fans behind hex mesh. Red rubber dampeners on the corners. Blue cutting mat underneath.
Build cost $3,000. Power $180/mo. Five clients pay $5,600 each per month.
Pause at 0:04 — red rubber dampeners still stuck to a former mining rig. Anthropic just billed his ex-team more this month.
Anthropic bills enterprises $28K/month for API at scale. He runs the same models on a $200 chassis from Alibaba. Break-even in 12 days.
Every abandoned mining rig just became a viable AI server. Anthropic doesn't see it coming.
Full setup in the video.
Knzo@0xKnzo
English
SOMEONE BUILT A NEW GENSHIN IMPACT WITH OPEN WORLD GAME IN 3 DAYS USING ONLY CLAUDE CODE
A few years ago, this required a studio, artists, Unity devs, environment designers and months of work
Now one person can use Hotcode to generate the core game: open world map, anime style character, movement, combat, quests, UI and playable prototype
Most people still think game development starts with learning 12 tools
That is old thinking
The new edge is not building every asset by hand. It is knowing how to direct AI tools until a game idea becomes playable fast
A solo creator with the right workflow can now test what used to cost a team
karpacho@karpachoq
English
Opus 5 just produced an unusually strong Rocket League style clone.
The result is not impressive only because it is playable. The car handling, arena structure, reflections, lighting, camera feel, and overall visual coherence are all well above what most one-shot game demos usually achieve.
What also stands out is the efficiency: this was done while consuming only 27% of a 5x Max subscription.
At least in game creation and 3D modeling, Opus 5 is starting to look like one of the strongest models in the world right now.
The playable version and the source code download will be in the comments.
Cc: @LLMJunky
English
If you still think benchmark scores tell the whole story, you haven't been paying attention.
The Hype compared Claude Opus 5, Fable 5, GPT-5.6 Sol, and Kimi K3 on the exact same procedural Three.js engineering tasks
Each model had to build:
• M4 Carbine
• Glock 18C
• Steyr TMP
GPT-5.6 Sol was the fastest and cheapest, Opus 5 delivered the best overall quality, Fable 5 came close but at a higher price, while Kimi K3 took the longest to finish
Price + runtime:
• GPT-5.6 Sol - $1.26 | 7.8 min
• Kimi K3 - $3.60 | 93.8 min
• Opus 5 - $6.22 | 24.8 min
• Fable 5 - $8.12 | 23.5 min
The interesting part wasn't who won, it was seeing how differently each model traded off quality, speed, and cost on the exact same prompts
thehype.@thehypedotnews
claude opus 5 vs fable 5 vs gpt 5.6 sol vs kimi k3 @AnthropicAI released claude opus 5 today. per the announcement, it's "a thoughtful and proactive model that comes close to the frontier intelligence of claude fable 5 at half the price." key facts from the release: • priced at $5/mtok input, $25/mtok output – unchanged from opus 4.8 • anthropic reports it as state-of-the-art on frontier-bench v0.1 (43.3%), gdpval-aa v2 (1861), arc-agi-3 (30.2%), zapier automationbench (26%) and osworld 2.0 (70.6%) • per the same release, it stays behind mythos 5 on cybersecurity, and behind fable 5 on parts of agentic coding (deepswe 68.8 vs sol's 72.7, frontiercode 53.4 vs fable's 53.5) • runs at max thinking effort by default; a fast mode is offered at ~2.5x speed for 2x the price. our runs used the standard setting our test – 3 prompts, single-file html, @threejs, fully procedural, no assets. each has a small fire button (web audio gunshot + muzzle flash + recoil + ejected brass) and a disassemble toggle that explodes the gun into labelled parts and rebuilds it: 1. 5.56 m4 carbine – collapsible stock, safe·semi·burst lower, quad-rail with r14–r28 panel numbers, aimpoint red-dot, vertical foregrip, folding bipod, a2 flash hider, breaks into 10 parts 2. glock 18c – select-fire machine pistol, ported 18c compensator slots, "glock 18c / austria 9x19" roll-marks, extended 33-round mag, field-strips into slide, ported barrel, recoil spring, frame, mag 3. steyr tmp – ribbed polymer housing, threaded barrel, integral forward vertical foregrip, canted translucent 30-round mag ran on @aimlapi results: - cost #1 gpt 5.6 sol – $1.26 #2 kimi k3 – $3.60 #3 opus 5 – $6.22 #4 fable 5 – $8.12 - tokens #1 gpt 5.6 sol – 36,554 #2 fable 5 – 120,203 #3 opus 5 – 137,093 #4 kimi k3 – 196,596 - lines of code #1 opus 5 – 2,738 #2 gpt 5.6 sol – 2,432 #3 kimi k3 – 1,907 #4 fable 5 – 1,764 - generation time #1 gpt 5.6 sol – 7.8 min #2 fable 5 – 23.5 min #3 opus 5 – 24.8 min #4 kimi k3 – 93.8 min observations: • opus 5 wrote the most code and shipped the best-looking guns – the two go together. 2,738 lines, almost no boilerplate, and it beat fable at 3/4 of fable's price. all 3 models render, all 3 disassemble reliably • it is the only model that respected the black-on-black problem. its studio env is a narrow overhead strip with a comment explaining why ("narrow + long beats big + square here"), and parts separate by roughness, not colour – slide 0.44, ribs 0.52, grip 0.86. that is why the lighting reads like a product shot • the steyr ribbing and m4 proportions are the best in the test. the ribs are real half-buried cylinders merged into one geometry, each catching a single specular line – not a bump map. thin barrel, slim quad-rail, correct 84cm scale, all computed from named stations, not eyeballed • price per 100 shipped lines – gpt sol $0.052, kimi $0.189, opus $0.227, fable $0.460. opus buys frontier-grade detail for half of fable's per-line rate opus 5's code quality: upsides: - it engineers instead of eyeballing. everything is built to real-world scale (m4 = 84cm), the rail teeth sit at the exact 1cm pitch a real picatinny rail uses, and the steyr's ribbing is actual 3d geometry catching a specular line, not a bump texture faking it - it caught a classic three.js trap the others didn't – the bevel on extruded shapes quietly inflates the silhouette, so it pre-corrected every surface constant so decals and ribs sit flush instead of floating or z-fighting downside: - one systemic bug: every left-side engraving renders mirrored. roll-marks, the safe·semi·burst selector, the mag numbers – all reversed. single cause, but it hits all three guns opus 5 is the strongest engineer in this field – it saw traps the others walked into and reasoned about physics constants, real-world scale, and sequences instead of guessing follow @thehypedotnews for 24/7 ai news, analysis and breakdowns
English
Asymet retweetledi
THIS $80 AI SYSTEM NEVER TAKES ITS EYES OFF THE TARGET.
Built with OpenCV, Arduino and two servo motors, it detects a face and automatically follows every movement in real time.
The same basic technology can power security cameras, search-and-rescue robots and autonomous defense systems.
Cheap prototype. Very serious implications.
x.com/asymetx/status…
Asymet@asymetx
English
HE'S LYING IN BED WATCHING PLANES CROSS HIS CEILING IN REAL TIME.
A guy from Austin built the setup himself, a Raspberry Pi, a cheap radio dongle, and a projector that turns a public aviation signal into a live map above his bed. Callsign, altitude, route, speed, updating as each plane passes over.
Claude glued the visualization together, turning raw signal data into the map projected overhead.
$200 spent.
Parts: Raspberry Pi 5, RTL-SDR ADS-B dongle, antenna, mini projector.
A comparable all-in-one ADS-B receiver retails for $799.
He built the core function for about a quarter of that price.
Picture it at night. Just the projector's glow and planes crossing the ceiling above the bed. Isn't that something?
Bober_smart@Bober_smart
English
A 20-year-old guy from South Korea created an AI girl and made $6,980 in just 18 days
He used Claude to generate both the images and short videos
Then he launched a TikTok account where she posted dance videos
Within a week, the account had already gained more than 61,000 followers
After seeing the engagement, he opened a Fanvue page for the character
> TikTok for reach
> Fanvue for monetization
In just 18 days, 634 people subscribed at $10.99/month, generating around $6,980 in revenue
AI keeps creating business models that barely existed a few years ago
Lyren_o@0xAIlabs
English
I created an AI version of my girlfriend to force her to dance on TikTok
She always wanted to be a blogger but was way too lazy to actually set up a camera.
So I used Claude Code and Kling 3.0 to build a fully automated pipeline that locked in her exact facial DNA.
Weeks went by with absolutely zero results. She even called me an idiot and said the whole project was a massive waste of time.
I just told her to trust the process.
Exactly one week later, a single video exploded to 10 million views.
Now she has thousands of new subscribers and an inbox full of paid brand offers, all without stepping in front of a real camera.
Should we just drift apart now?
Spivach@0xSpivach
English
CLAUDE DESIGN VS FRAMER, GAVE BOTH THE SAME PROMPT, HERE'S WHERE THEY ACTUALLY SPLIT
same brief, same amount of detail, completely different strengths show up by the second screen
Claude Design nails structure and logic fast, it reasons through the layout like a dev thinking in components, but the visual polish still needs a second pass
Framer comes out the gate with the finished look, typography, spacing, motion, all dialed in because it's a design-first tool wearing an AI layer, not the other way around
the real difference isn't which one "wins," it's what stage you're at: concepting and structure → Claude, ready-to-publish and pixel-perfect → Framer
most people pick one and force it to do both jobs, that's where the AI slop comes from
which one's actually in your stack right now?
Kenny1st@0xKenny1st
English
She makes over $43,000 a month from 15-second bedroom dance clips.
Most creators still think they're competing with other influencers.
They're already competing with AI.
She isn't real.
Every frame, every expression, and every movement is generated, refined, and optimized by AI.
Behind the scenes, a 23-year-old student from Texas runs the entire operation from his bedroom using Python scripts, an RTX GPU, and a fully automated content pipeline.
Trending audio is detected automatically, videos are generated, edited, and distributed across more than 20 accounts with almost no manual work.
Just systems running 24/7.
The result?
25M+ weekly views.
$43,000+ every month.
AI isn't replacing creators.
It's creating entirely new ones.
kerazcity@0xkerazcity
English
An American creator turned herself into a Korean AI model — and made $128,345 from it.
That’s roughly $10,695 per month if the total covers one year.
With Claude helping her create the videos, she launched a blog that pulled 3,000,000+ views.
One woman. Multiple identities. A six-figure AI business.
x.com/asymetx/status…
Asymet@asymetx
English