cokk_b

[🌺Sai🌺 @Sai60012]

No one is greater than Om.

o
m
o
m
o
m
o
m
good morning my dear friend 💗
o
m
o
m
o
m
o
m
Sabhi ko Radhe Radhe🪷
o
m
o
m
o
m
o
m
Om namah shivaya 🔱
o
m
o
m
o
m
o
m
Har Har Mahadev
o
m
o
m
o
m
o
m @grok

[NHKニュース @nhk_news]

通常診療を再開した熊本県八代市の「熊本総合病院」

今回の熊本地震の瞬間、手術室の緊迫した様子が病院のカメラに映っていました↓
https://t.co/TS7biPg7wQ

[cokk_b @cokk_b]

Autumn mornings are a whole vibe | Tịnh

[Asmavia Chaudhary🇵🇰 @AsmaviaChaudhry]

I❤️
محمد ﷺ
MUHAMMAD

I❤️
محمد ﷺ
MUHAMMAD

I❤️
محمد ﷺ
MUHAMMAD

I❤️
محمد ﷺ
MUHAMMAD

I❤️
محمد ﷺ
MUHAMMAD

I❤️
محمد ﷺ
MUHAMMAD

I❤️
محمد ﷺ
MUHAMMAD

I❤️
محمد ﷺ
MUHAMMAD

I❤️
محمد ﷺ
MUHAMMAD

I❤️
محمد ﷺ
MUHAMMAD

I❤️
محمد ﷺ
MUHAMMAD

I❤️
محمد ﷺ
MUHAMMAD

[cokk_b @cokk_b]

My cat sneezed and scared themselves and I've never felt more seen | Trai

[Frezz @Frezzwnie]

We’re living in the age of the anti-brain.

Every day brings more articles, videos, other people’s ideas, and random thoughts that appear for a few seconds before disappearing. Everything ends up scattered across notes, screenshots, and “read later” bookmarks.

Then one day I open Obsidian and realize something:

I can’t find any of it.

Thousands of notes.
Zero structure.

I can barely remember what I had for lunch yesterday-let alone an idea I wrote down two months ago.

The solution turned out to be ridiculously simple:

Stop trying to keep everything in my head.
Move my brain into Obsidian.

Here’s how it works:

I build a single knowledge graph of my notes and AI sessions. It’s not a folder of files-it’s my external working memory.

Claude Code, Codex, OpenClaw, and Hermes read directly from that knowledge base. I never have to re-explain context. Every session continues exactly where the last one ended, with access to everything that came before.

Everything I study-YouTube videos, official documentation, GitHub repositories, websites-flows into the wiki automatically every day. No manual copy-pasting.

The system filters itself. Valuable information stays. Noise disappears.

The best analogy I’ve found:

Imagine bringing books into a library.

A librarian puts each one on the right shelf, throws away anything that isn’t actually a book, and the next time you ask for something, finds it in seconds
like a Daiso employee who somehow knows where every single item in the store is.

Except this librarian works 24/7.

It never gets tired.

And it serves four AI agents simultaneously.

Crazy world we’re building.

[cokk_b @cokk_b]

Started tracking my workouts and the progress is real | Lô

[Makkah 🕋 @Harmai786]

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah
♥️ I LOVE ALLAH♥️

[cokk_b @cokk_b]

I left the house without brushing my hair and nobody said a thing - Hường N. Tôn

[SCREAM @scream_crypto]

Why build a $10,000+ local AI rig if a $600 Mac Mini types at the exact same speed?

Looking at this beast Quad setup with dual NVIDIA Sparks and ASUS nodes, most people see an expensive flex. If you only measure text generation, they are right. In a basic chat, a cheap Mac Mini and a high-end enterprise node feel identical.

But writing speed is a decoy.

The real bottleneck is prefill—how fast the system reads giant local codebases, RAG pipelines, and agent histories before typing the first word.

While the Mac Mini reads context at 564 t/s, a single DGX Spark obliterates it at 2,107 t/s. When you cluster multiple nodes together to run local autonomous agents like NVIDIA NemoClaw, that processing power scales into a completely different dimension.

The full teardown below breaks down exactly what to choose for your specific tasks, so go ahead and read it.

[bonduelle @bonduelleioat]

The era of writing a good prompt and sitting next to ChatGPT is over.

That chapter has ended. While most people are still working in a “human-in-the-loop” mode, modern agents like GPT-5.6 Sol can now handle hours of work completely autonomously.

Those who learn to control them first will gain an advantage that everyone else will struggle to catch.

But there’s one problem.

If you simply hit Run without the right guardrails, you could wake up to an empty account balance, a broken project, or a massive token bill.

The difference between a professional AI agent and a “digital vandal” is not the model itself - it’s the boundaries you build around it.

The true power of an agent is not its ability to “think” for hours. It’s the architecture that refuses to trust the model at face value.

An orchestrator should make the decisions, lower-cost specialized agents should handle repetitive tasks, and an independent verifier should evaluate the output against strict success criteria - not whether it merely “looks good.”

In practice, Sol is prone to what many call agentic drift: it can keep optimizing a task until it eventually starts breaking its own work.

That’s why your job is no longer to supervise every step.

Your job is to specify the desired outcome and build a system where every tool call, every code change, and every decision is constrained by sandbox environments, automated tests, and hard budget limits.

If your agent doesn’t have a built-in “stop command” that automatically rejects any result that fails tests, exceeds code limits, or goes beyond its budget, then you’re not controlling the system.

You’re simply hoping for the best.

When the architects behind these systems say that “the easy part is over,” this is exactly what they mean.

The future is no longer about smarter chatbots.

It’s about autonomous software factories, where reliability isn’t measured by benchmark scores but by your ability to close your laptop, let the system work through the night, and wake up to results proven by tests - not by convincing promises.

[cokk_b @cokk_b]

Anyone else feel like May has its own personality - Khải T. Tạ

[KalqiX @kalqix]

The first AMM live on KalqiX.

Trustless order book execution, running underneath an app traders already use. No rebuild, no bridge, no custody handoff.

Orders settle atomically on Avail Atomic by @AvailProject.

QuickSwap is the first. Not the last.

[fares . sh 🇵🇸 @fares__2001]

Unfortunately, no one cares 😭

︎
︎

︎

︎

︎

︎
︎1.6 m

︎

︎
︎

we’ll remind you
︎

︎

︎

︎

︎
︎

︎

︎

︎

︎
︎

︎
︎

︎
︎

︎

︎

︎
︎

︎
that
︎

︎

︎

︎

︎

︎
︎

︎

︎

︎

I desperately need TREATMENT for my baby😭 💔.,

https://t.co/VW7YnQfJX1

[cokk_b @cokk_b]

The cat judged me for eating cereal for dinner and I have it on video. - Chung N. Vũ

[RevenueCat @RevenueCat]

Shipaton is back. Submissions open August 1 – September 30.

14 categories.
2 months to ship.
$500k+ in prizes.

Plus, your app on a giant billboard in Times Square.

[cokk_b @cokk_b]

The accuracy though

[cokk_b @cokk_b]

Missing the mountains a lot today. - Ngộ T. Liễu

[Defn Pak @DefnPak]

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah

Allah
♥️ I LOVE ALLAH♥️

[cokk_b @cokk_b]

@ariyoshihiroiki Well said