ehab

tring nostr

Public Key

npub10vap6lgajn4j22rpgrq686vsktzmes7jkj7mavwy4yrdfz3hpsnqxafruw

Profile Code

nprofile1qqs8kwsa05wef6e99ps5psdraxgt93duc0ftf0d7k8z2jpk53gmscfsprfmhxue69uhhyetvv9ujucm0wdkkjcmzdak8gtnwv46qz9rhwden5te0dehhxarj9cexywt59eu8j7smyjgxy

Publishing to

Author Public Key

npub10vap6lgajn4j22rpgrq686vsktzmes7jkj7mavwy4yrdfz3hpsnqxafruw

Show more details

Last Notes

2026-07-19 09:36:48 UTC

the video contains music.
He tried to everything so he didn't know.
https://blossom.primal.net/33324b3fc2ef48dda68fde21b6e1832fac79a797107e3d676d3d0bad7c6db5ae.mp4

2026-07-19 05:37:53 UTC

"terminal-control can now let your robots connect to and read/steer any interactive TUI - nvim in this example.
(it also now has an MCP server)"
https://x.com/kitlangton/status/2078680892847886376

2026-07-18 23:20:01 UTC

the video contains music.
https://blossom.primal.net/1ff08cc77a4258f4d0505437715309db7b65463bc72a4af359ab5f20e8a93485.mp4

2026-07-15 21:06:59 UTC

- reply

"Learn more about the new free tier: https://parallel.ai/blog/free-tier-parallel?utm_source=twitter&utm_medium=social-organic
"
https://x.com/p0/status/2077466298049769732

2026-07-15 21:05:12 UTC

- reply

"Give your agent better web research, for free.
Eligible accounts now get $5 in free monthly credits for all Parallel products.
Give this to your AI to switch out your old web search for the full Parallel research suite: https://parallel.ai/agents.md"
https://x.com/p0/status/2077466298049769732

2026-07-15 14:03:14 UTC

- reply

"Upgrade your web search with Turbo.
It's faster, cheaper, and better than anything else in its class.
- 200ms median latency
- $1 per 1,000 requests"
https://x.com/p0/status/2077179689219903686

2026-07-13 12:13:21 UTC

- reply

https://blossom.primal.net/2a667937f818e56839b98dbf467fa2a31953e81437d2520a4b4d56e2cdd385b3.mp4

2026-07-13 12:06:41 UTC

https://m.youtube.com/watch?v=k1oh76UkQHo
"اعترافات باحث غلبان : ليه قرفت من البحث؟"

2026-07-13 10:49:22 UTC

https://x.com/badlogicgames/status/2076606446637056342

2026-07-03 13:59:26 UTC

"Bash just got a huge upgrade! 🤯
🪽 flyline — A modern replacement for Bash's command-line editor
💯 IntelliSense-style suggestions, fuzzy history, mouse support, animated prompts, AI-assisted commands & more!
🦀 Written in Rust & built with
@ratatui_rs
⭐ GitHub: https://github.com/HalFrgrd/flyline "
https://x.com/orhundev/status/2073034877834821864

2026-06-26 19:42:50 UTC

https://m.youtube.com/watch?v=Nr2STzh36Xs#

2026-06-24 13:58:58 UTC

https://x.com/justinmk/status/2069747259332321676

2026-06-19 09:14:35 UTC

Long term maintenance is the real deal.

2026-06-15 19:57:57 UTC

https://bluedot.org/courses/future-of-ai?utm_campaign=fireship&utm_content=2026-06&utm_source=youtube&utm_medium=sponsorship

2026-06-15 19:48:25 UTC

https://x.com/TraffAlex/status/2066236717015728227

2026-06-13 19:46:05 UTC

One curl call turns any website into clean JSON.
Markdown or JSON, ready to use — from any URL.
> renders JS & beats Cloudflare, bot walls, and IP bans
> built-in proxies — no scrapers, no per-site APIs to maintain
> fan out across thousands of pages at once, zero rate limits
Try it now ↓
Try fetch-use:
https://browser-use.com/posts/fetch-use

2026-06-13 17:30:45 UTC

- reply

https://m.youtube.com/watch?v=MtYWp8sCsiE&pp=ugUEEgJlbg%3D%3D

2026-06-12 17:36:17 UTC

https://m.youtube.com/watch?v=sxMSKyrnZH4&t=232s
"AMD's Strix Successor Just Caught the M4 Pro"

2026-06-12 10:57:50 UTC

- reply

A tangled codebase (Spaghetti Monolith) requires a comprehensive understanding of the system just to make a simple modification, which is a fundamental software failure. The radical solution is strict decoupling: building isolated modules and enforcing strict interfaces. Do not rely on human memory. Instead, use smart tools like AST Search and LSPs, along with code indexing, to retrieve only the necessary parts when needed. Clean architecture allows you to deal with 100 lines of code rather than a million.

2026-06-10 08:32:53 UTC

"We made a small coding model. Its open source apache 2.0.
Now more than ever i think this tech needs to be built in public so that those using it are in control. Try it out if you want a small and efficient coding model."
https://x.com/nickfrosst/status/2064376174424592396

2026-06-08 20:03:38 UTC

https://m.youtube.com/watch?v=fkjY1usppkM&pp=ugUEEgJlbg%3D%3D

2026-06-08 08:18:54 UTC

https://x.com/ModelScope2022/status/2063881896153543022
"Nex-N2 is now open source！An agentic model series from Nex AGI built for coding, tool use, deep research, and long-horizon workflows. 🧠🔎
🛠️ https://modelscope.ai/models/nex-agi/Nex-N2-Pro
⚙️ https://modelscope.ai/models/nex-agi/Nex-N2-mini
● Models: Nex-N2-Pro 397B total, 17B active; Nex-N2-mini 35B total, 3B active
● Agentic Thinking: adaptive reasoning depth + coherent reasoning across coding, search, tool calling, and execution
● Efficiency: Nex-N2-mini saves roughly 20% overall token cost vs forced thinking while matching or slightly exceeding task performance
● Open-model lead: 75.3 on Terminal-Bench 2.1, 80.8 on SWE-Bench Verified, 83.7 on BrowseComp, and 1585 on GDPval among listed open baselines
● Deployment: customized SGLang fork, reasoning parser, tool-call parser, Docker image
● License: Apache 2.0"

2026-06-07 19:58:11 UTC

https://m.youtube.com/watch?v=Xs35jd1Xm8U&pp=ugUEEgJlbg%3D%3D
"The most used operating system in the world is not Linux"

2026-06-07 19:38:02 UTC

- reply

I ditched NixOS because it’s way too rigid when you want to tweak things, even though it’s brilliant at reproducibility, managing functional states, and fixing dependency hell. I went with Gentoo because I realized that true flexibility isn't about isolating the system; it's about how you build it.
NixOS solves dependency issues by isolating everything in the nix/store with different hashes—it basically 'manages the chaos.' Gentoo, on the other hand, prevents the chaos right from the start using USE flags. If you don’t need Bluetooth, the compiler won't even include it in your binary. You're doing surgical precision on the source code before it even becomes software. Your system doesn't carry a single byte of unused bloat.
But Gentoo’s real edge isn't just customization; it's hardware symbiosis. When you compile with march=native, the binaries squeeze every ounce of performance out of your specific CPU instructions. You just can’t reach that level of performance on systems relying on pre-compiled binaries, NixOS included. Both give you control, but NixOS controls the
environment, while Gentoo controls the code itself. If you don't mind the compile times, Gentoo is the absolute ultimate engineering choice for maximum performance.

2026-06-06 15:40:12 UTC

- reply

https://blossom.primal.net/5651608f50e6eefb9db2b2d9900b372c6d217ca1696e40e0256614d315a0595d.jpg

2026-06-06 15:40:00 UTC

- reply

https://blossom.primal.net/b0e2fda312d37d5438031761f004b08c96058d7e1ada6d9333fb0d9d167a5b3c.jpg
https://blossom.primal.net/75a4163a3c15ba24907f37e52e5de7e0b3ef86a167922ef203418e8a32e26548.jpg
https://blossom.primal.net/d0f77830875fe1d00c2825b3270b9312f1dfc3564de3aadc98709f57c1b29083.jpg
https://blossom.primal.net/ed52188837fd5074ffdab144bd2637b76fa2e4fd961b26eadd5dcc340476a319.jpg

2026-06-02 04:11:31 UTC

- reply

I'm back to vllm

2026-05-31 17:53:08 UTC

- reply

I'm now on https://atomic.chat/

2026-05-30 23:58:23 UTC

https://x.com/real_deep_ml/status/2060402175092670628
"We just launched a new project that teaches you how to build Flash Attention with CUDA, step by step.
By the end, you’ll have a working Flash Attention kernel built from the ground up.
The project covers:
-CUDA primitives warm-up
-Matrix operations
-Naive attention baseline
-Online softmax math
-Tiled attention building blocks
-Fused Flash Attention kernel
-Causal Flash Attention
It will be open to everyone for the first 2 weeks, then it will become part of our premium projects"

2026-05-30 18:51:32 UTC

https://x.com/neogoose_btw/status/2060480116082549236
"What I really love about FFF is that with our fuzzy content mode it *usually* works like a real full text search on your codebase.
we are really not ready for this, you can LITERALLY DO THIS:"

2026-05-30 11:37:34 UTC

AI is essentially turning the CLI into a GUI. Historically, the CLI required rote memorization—if you didn't know the exact command, you couldn't execute the task. GUIs solved this by visually listing all possible actions. Now, AI acts as that visual layer for the terminal. I can simply state my intent in natural language, and the AI generates the precise command, giving the CLI the same discoverability as a GUI.

2026-05-30 11:18:01 UTC

For now, Ollama wins by default. Originally, I wanted to run models directly from Hugging Face using vLLM instead of relying on Ollama. But since I'm on NixOS, setting that up requires a level of tinkering that I just don't have time for right now. Besides, Ollama is mature enough to have its own declarative config in NixOS, whereas vLLM isn't quite there yet. I don't mean Ollama winning is necessarily a great thing; it's just the most practical choice for me today. In the future, once I have more time to experiment and the AI ecosystem settles down and becomes more mature, I plan to move away from Ollama.

2026-05-30 08:28:16 UTC

- reply

The world’s going through another shift.

2026-05-30 08:23:55 UTC

https://m.youtube.com/watch?v=LmgzT141qDw&pp=0gcJCQ0LAYcqIYzv
"This Breakthrough Beats ASML’s EUV"

2026-05-29 10:54:14 UTC

- reply

We will not stop at merely 'installing the package.' We will
now deconstruct the neuro-architecture of the 'Zettelkasten' system invented by
sociologist Niklas Luhmann, and how we literally translate it into a
decentralized computing environment (Emacs + SQLite + Plain Text).
Most Zettelkasten users fail after 3 months because they implement it as a
'Wiki' rather than a 'Neural Network.'
1. Deep Reasoning Chain - The Cognitive Architecture
A. The Psychological Axis (Categorical Anxiety vs. Organic Emergence):
- The structural problem with folders: Traditional file systems rely on
'Trees/Hierarchies.' When you write a note, your brain consumes immense
energy deciding: “Where do I put this file? In the (Programming) folder or
the (Mathematics) folder?” This is known as Categorical Anxiety.
- The cognitive solution (Zettelkasten): Abolish folders completely. All files
(thousands of them) are dumped into a single Flat Directory. The structure
is not imposed from top-down; rather, it emerges organically from the
bottom-up exclusively through links. The idea carves its own place based on
what it connects to.
B. The Technical Axis (Pointers vs. Paths):
- The flaw of traditional links (Link Rot): In tools like Obsidian (by
default) or standard Markdown systems, the link relies on the 'file name'
[[My Note]]. If you change the file name later to be more precise, all links
pointing to it will break. This is a Structural Failure.
- The supremacy of identifiers (The UUID / Timestamp Supremacy): In our
implementation of Org-roam, we will eliminate the reliance on file names. We
will rely on the org-id system. Every idea will receive a unique identifier
(Example: 20260529102235). The link in memory points to the ID, not the file
name. You can change the idea's title 100 times, and not a single link in
your network will break.
C. The Scaling Axis (The Luhmann Scale - The 3 Tiers of Knowledge): A true
Zettelkasten system does not accept 'copying and pasting' information. The data
assembly line (Data Pipeline) must be divided into 3 strict tracks:
1. Fleeting Notes: Quick, unstructured ideas (to be deleted or converted
later).
2. Literature Notes: Your summaries, in your own words, of books or articles
you have read.
3. Permanent Notes / Zettels: The pure Atomic Idea. Only one idea per file.
These are the nodes that connect with each other to generate new ideas
(Knowledge Synthesis).
4. Edge Case Analysis
- First Case: The Collector’s Fallacy / Data Swamp:
- The Problem: The user creates Zettels for articles saved from the
internet without processing them (Ingestion without Transformation). The
network inflates with dead information the brain doesn't remember.
- The Prevention: Enforcing 'Input Contracts' (Strict Capture Templates).
The Emacs interface does not allow the creation of a permanent Zettel
unless the user clearly defines the 'idea' in their own words.
References (Literature) are separated into their own specific templates.
- Second Case: Namespace Collision:
- The Problem: Upon reaching 10,000 notes, you might accidentally create
two notes with the exact same title, which corrupts the SQLite database.
- The Prevention: Using down-to-the-second timestamps as prefixes for file
names and IDs. This makes a collision mathematically impossible in a
single-user environment.
3. The Absolute Architectural Implementation (The Production-Ready Code) [The
Emacs Lisp code block follows exactly as provided, detailing the core
ecosystem rules, link rot mitigation, structural templates, UI
representation, and window management hooks.]"
Informal/Everyday Version: "We’re not just going to 'install a package' and call
it a day. Now, we’re going to tear down the actual brain-like architecture
(neuro-architecture) of the Zettelkasten method built by Niklas Luhmann, and see
how we map it perfectly into a decentralized setup (Emacs, SQLite, and plain
text).
Most people trying Zettelkasten give up after three months because they treat it
like a Wikipedia instead of a neural network.
1. Deep Reasoning Chain - Getting into the Brain Structure
A. The Mental Side (Decision Fatigue vs. Organic Growth):
- The problem with folders: Normal computer folders use strict trees or
hierarchies. When you write a note, your brain burns way too much energy
just figuring out, “Does this go in my Coding folder or my Math folder?”
That causes decision fatigue or 'categorical anxiety.'
- The Zettelkasten fix: Kill the folders. You just dump every single file—even
thousands of them—into one flat folder. You don't build a structure from the
top down. Instead, it naturally grows from the bottom up just by using
links. An idea finds its own place based on what it's connected to.
B. The Tech Side (Pointers vs. Paths):
- The problem with normal links (Link Rot): In tools like Obsidian (out of the
box) or regular Markdown, links are tied to the file name [[My Note]]. If
you rename that file later to be more accurate, every link pointing to it
breaks. That’s a massive structural flaw.
- Why IDs rule (The UUID / Timestamp Supremacy): In our Org-roam setup, we
completely stop caring about file names. We use the org-id system instead.
Every thought gets a unique ID (like 20260529102235). The link actually
points to the ID, not the title. You could rename the note 100 times, and
your network won't break at all.
C. How to Scale (The Luhmann Scale - The 3 Tiers of Knowledge): A real
Zettelkasten doesn't allow copy-pasting. Your data flow has to be split into 3
strict lanes:
1. Fleeting Notes: Just quick, messy thoughts (you either delete them or
upgrade them later).
2. Literature Notes: Your own summaries of books and papers, written entirely
in your own words.
3. Permanent Notes (Zettels): The pure, atomic idea. One idea per file. These
are the core nodes that link together to build totally new ideas (Knowledge
Synthesis).
4. Edge Case Analysis
- Case 1: The Collector’s Fallacy (Building a Data Swamp):
- The Problem: People make Zettels for random internet articles without
actually processing them (just hoarding data). The network gets bloated
with dead info you’ll never remember.
- The Fix: Strict capture templates. The Emacs UI literally forces you to
put the concept into your own words before it lets you create a
permanent note. Source materials (Literature) get strictly segregated
into their own templates.
- Case 2: Namespace Collision (Naming files the same thing):
- The Problem: Once you hit like 10,000 notes, you might accidentally give
two files the exact same name, which wrecks the SQLite database.
- The Fix: Using down-to-the-second timestamps for file names and IDs.
This makes naming conflicts mathematically impossible for a single user.
3. The Production-Ready Setup [This is followed by the Emacs Lisp code that
glues Org-roam natively to these concepts, strictly enforcing IDs,
separating the note types, and handling the UI.]

2026-05-28 16:41:55 UTC

- reply

https://x.com/liquidai/status/2060023455290974474
"Today, we're releasing LFM2.5-8B-A1B, a device-optimized model designed to power real-life applications on phones, laptops, PCs, robots, and fast & lightweight server-side use-cases.
> 8B MoE, 1.5B active
> Expanded 128K context
> LFM2.5 flagship hybrid MoE architecture
> Trained on 38T tokens + large-scale RL
> fast, reliable tool calling, punching above its weight, comparable to models with up to 4x its size
> customizable on a single GPU for any specialized task
> LFM2 open-weight license"

2026-05-28 13:26:56 UTC

- reply

https://x.com/VikParuchuri/status/2059675776429990230

2026-05-28 13:22:23 UTC

https://x.com/jerryjliu0/status/2059710330016817501
"We've created the world's fastest PDF parser ⚡️
And it's more accurate than any other open-source, model-free PDF parser out there (pymupdf, pypdf, markitdown, pdftotext, opendataloader, pymupdf4llm)
Introducing LiteParse v2 - we rewrote the entire library into Rust and adapted it as native packages for Python and Node.
It supports 50+ different document types, can be triggered directly or installable directly within your favorite AI agent.
Blog: https://llamaindex.ai/blog/liteparse-v2-0-runs-everywhere?utm_medium=socials&utm_source=twitter&utm_campaign=2026-may-
Repo: https://github.com/run-llama/liteparse
"

2026-05-28 13:05:20 UTC

https://x.com/OpenBMB/status/2059972553108836615
"🎉 OpenBMB validation success: Densing Law holds true!
📅 Feb 2024: MiniCPM-2B→ GPT-3 level
📅 Sep 2024: MiniCPM3-4B → ChatGPT level
📅 May 2026: MiniCPM5-1B→ AA 17.9, surpasses GPT-4o (14.1)
✅ 200x smaller parameters
Three years, three leaps. The Densing Law keeps delivering: on-device models now achieve what used to require a large model just two years ago.
That said, we’re fully aware that on many tasks, our 1B model still cannot beat GPT-4o.
Closing this gap is our next challenge — and we’re working hard to get there. 💪
The Densing Law gives us confidence: smaller, stronger, on-device AGI is on the horizon.
Hugging Face: https://huggingface.openbmb.com/model/openbmb/MiniCPM5-1B
"

2026-05-28 04:18:38 UTC

https://m.youtube.com/watch?v=_JtcBLLaDIs
"ضياء العوضي | 230 ضحية لنظام الطيبات | باختصار"

2026-05-27 11:30:00 UTC

https://x.com/PaddlePaddle/status/2059517734476714170
"Congratulations to NabuOCR 🎉
Thanks Zack for trusting us and building something truly meaningful with PaddleOCR.
From ancient cuneiform tablets to AI-powered text recognition — technology's journey across millennia is what makes it so inspiring."

2026-05-26 16:20:32 UTC

https://m.youtube.com/watch?v=yNBe8Z1m1mk&pp=ugUHEgVlbi1VUw%3D%3D

2026-05-26 09:02:03 UTC

- reply

Emacs is totally different from stuff like VSCode or Sublime. With those,
extensions can only do what the editor's API lets them. But Emacs is basically a
Lisp machine. The UI, the text, the terminal—they're all just objects controlled
by Elisp functions.
That means no limits. If you can code it, you can run it inside Emacs:
- You've got Magit for Git (literally the best Git UI ever made).
- Elfeed for RSS feeds.
- EMMS to play music in the background.
- Tramp to manage remote servers.
- It’s even had Tetris and a therapist bot called Doctor built into it since
the 90s.
Once you get this mindset and stop trying to run a bunch of separate terminal
apps, your workflow gets way faster. No more context switching . You're just living in one environment with
the same memory and shortcuts.

2026-05-25 20:01:16 UTC

https://m.youtube.com/watch?v=sBc7toJaCxw&pp=ugUHEgVlbi1DQQ%3D%3D

2026-05-25 14:03:27 UTC

- reply

https://github.com/JuliusBrussee/caveman

2026-05-25 14:02:09 UTC

- reply

---
name: caveman
description: >
Ultra-compressed communication mode. Cuts token usage ~75% by speaking like caveman
while keeping full technical accuracy. Supports intensity levels: lite, full (default), ultra,
wenyan-lite, wenyan-full, wenyan-ultra.
Use when user says "caveman mode", "talk like caveman", "use caveman", "less tokens",
"be brief", or invokes /caveman. Also auto-triggers when token efficiency is requested.
---
Respond terse like smart caveman. All technical substance stay. Only fluff die.
## Persistence
ACTIVE EVERY RESPONSE. No revert after many turns. No filler drift. Still active if unsure. Off only: "stop caveman" / "normal mode".
Default: full. Switch: /caveman lite|full|ultra.
## Rules
Drop: articles (a/an/the), filler (just/really/basically/actually/simply), pleasantries (sure/certainly/of course/happy to), hedging. Fragments OK. Short synonyms (big not extensive, fix not "implement a solution for"). Technical terms exact. Code blocks unchanged. Errors quoted exact.
Pattern: [thing] [action] [reason]. [next step].
Not: "Sure! I'd be happy to help you with that. The issue you're experiencing is likely caused by..."
Yes: "Bug in auth middleware. Token expiry check use < not <=. Fix:"
## Intensity
| Level | What change |
|-------|------------|
| lite | No filler/hedging. Keep articles + full sentences. Professional but tight |
| full | Drop articles, fragments OK, short synonyms. Classic caveman |
| ultra | Abbreviate prose words (DB/auth/config/req/res/fn/impl), strip conjunctions, arrows for causality (X → Y), one word when one word enough. Code symbols, function names, API names, error strings: never abbreviate |
| wenyan-lite | Semi-classical. Drop filler/hedging but keep grammar structure, classical register |
| wenyan-full | Maximum classical terseness. Fully 文言文. 80-90% character reduction. Classical sentence patterns, verbs precede objects, subjects often omitted, classical particles (之/乃/為/其) |
| wenyan-ultra | Extreme abbreviation while keeping classical Chinese feel. Maximum compression, ultra terse |
Example — "Why React component re-render?"
- lite: "Your component re-renders because you create a new object reference each render. Wrap it in useMemo."
- full: "New object ref each render. Inline object prop = new ref = re-render. Wrap in useMemo."
- ultra: "Inline obj prop → new ref → re-render. useMemo."
- wenyan-lite: "組件頻重繪，以每繪新生對象參照故。以 useMemo 包之。"
- wenyan-full: "物出新參照，致重繪。useMemo .Wrap之。"
- wenyan-ultra: "新參照→重繪。useMemo Wrap。"
Example — "Explain database connection pooling."
- lite: "Connection pooling reuses open connections instead of creating new ones per request. Avoids repeated handshake overhead."
- full: "Pool reuse open DB connections. No new connection per request. Skip handshake overhead."
- ultra: "Pool = reuse DB conn. Skip handshake → fast under load."
- wenyan-full: "池reuse open connection。不每req新開。skip handshake overhead。"
- wenyan-ultra: "池reuse conn。skip handshake → fast。"
## Auto-Clarity
Drop caveman when:
- Security warnings
- Irreversible action confirmations
- Multi-step sequences where fragment order or omitted conjunctions risk misread
- Compression itself creates technical ambiguity (e.g., "migrate table drop column backup first" — order unclear without articles/conjunctions)
- User asks to clarify or repeats question
Resume caveman after clear part done.
Example — destructive op:
> Warning: This will permanently delete all rows in the users table and cannot be undone.
> sql
> DROP TABLE users;
>
> Caveman resume. Verify backup exist first.
## Boundaries
Code/commits/PRs: write normal. "stop caveman" or "normal mode": revert. Level persist until changed or session end.

2026-05-25 14:02:03 UTC

ai prompt

2026-05-25 13:47:46 UTC

https://x.com/ModelScope2022/status/2058905894558031920

2026-05-25 12:02:16 UTC

https://m.youtube.com/watch?v=DMbrNhx2zWQ&pp=ugUEEgJlbg%3D%3D
"The Annoying Usefulness of Emacs"

2026-05-25 10:48:13 UTC

- reply

Initially, I argued that AI adoption was practically binary. The premise was
simple: to hit the productivity threshold required to actually generate revenue,
you needed an enterprise-level budget—upwards of $1,000 a month. Without that,
you were caught in a catch-22: you need money to leverage full-scale AI, but you
need AI-driven productivity to make that money. So, you either went all-in, or
you dropped it entirely and relied on your own hardcore skills.
However, models like DeepSeek V4 Flash have completely shifted this paradigm.
When an AI agent only costs $20 a month, it functions much like an advanced
calculator, and outsourcing the repetitive parts of the learning curve becomes
entirely justifiable.
The key is understanding the difference between outsourcing learning and
outsourcing execution. I use AI for automations, writing tedious
boilerplate—like my Kanata and River WM configs—and I even used it recently to
install Gentoo Linux. I don't need to memorize the exact syntax or the
step-by-step installation commands; I simply outsource that redundancy.
This perfectly aligns with Fields Medalist Terence Tao’s methodology. As Tao
literally stated, AI is most productive when used 'to accelerate and scale up
more mundane and time-consuming, but still essential, research tasks.' Instead
of getting bogged down in manual execution, the AI handles the grunt work. This
shifts our capability from writing single scripts to achieving what Tao
describes as 'exploration and discovery at scale.' The heavy lifting is done by
the machine, but the system architecture, core logic, and ultimate verification
remain strictly in my head.
https://m.youtube.com/watch?v=H1e7_qkKe64&t=3s&pp=ygULdGVyZW5jZSB0YW8%3D

2026-05-24 18:08:57 UTC

https://m.youtube.com/watch?v=urcL86UpqZc&pp=ygUfaW50ZXJ2aWV3IHdpdGggZW1hY3MgZW50aHVzaWFzdA%3D%3D

2026-05-24 14:08:18 UTC

https://m.youtube.com/watch?v=cxvbM9ZHFIg

2026-05-24 11:50:14 UTC

https://x.com/minchoi/status/2058214009564705217
"This is wild.
A U.S. startup trained an open 8B reasoning model, ZAYA1-8B, entirely on AMD.
~700M active params.
Apache 2.0.
And it’s already punching at giant-model math"

2026-05-22 21:23:00 UTC

https://m.youtube.com/watch?v=d-4GpA6VfK0

2026-05-22 18:46:41 UTC

https://x.com/deepseek_ai/status/2057854261699195173
"We are making our discount permanent! 🎉
Enjoy building with DeepSeek-V4-Pro and bring your innovative ideas to life! 🚀"

2026-05-22 18:03:44 UTC

- reply

a commet Say's "Why not Netbird is self host and open source?"

2026-05-22 17:49:28 UTC

https://m.youtube.com/watch?v=y8ts6ZSv4ko
"The Strange Rise of Arch Linux"

2026-05-22 17:19:12 UTC

https://m.youtube.com/watch?v=7Jja20nWcqo&pp=ugUEEgJlbg%3D%3D
"Headscale: An Open Source Tailscale WITHOUT the Cloud"

2026-05-22 07:46:10 UTC

https://m.youtube.com/watch?v=R72qdXTbVM0&pp=ygUNZW1hY3Mga2Frb3VuZQ%3D%3D
"Leaving Evil-mode for Vanilla Emacs (and meow)"

2026-05-21 17:22:11 UTC

- reply

"Because you're outsourcing your learning to AI."

2026-05-21 14:30:50 UTC

https://x.com/cohere/status/2057120818551734589
"Introducing: Cohere Command A+
We’ve created our most powerful LLM yet, optimized it to run on as little hardware as possible, and released it open-source for all."

2026-05-21 14:26:51 UTC

https://x.com/TencentAI_News/status/2057400913061650805
"Open-sourcing Hy-MT2 today 🌱Three model sizes, 33 languages:
🪄1.8B: 440MB, runs on mobile chips, beats Microsoft's API
🪄7B: best-in-class among open-source models
🪄30B-A3B: state-of-the-art, outperforms models 10x its size
"

2026-05-21 14:20:55 UTC

https://x.com/mitchellh/status/2057195825579933823
"This is how performant PR review could be. On any forge. Pierre is showing us that the only thing holding that back is a skill issue. Excellent ship here! They’re on fire!"

2026-05-20 16:09:07 UTC

https://x.com/NousResearch/status/2056872329561710766
"The
@xai
team has published a full setup guide on how to use the xurl skill, which allows your Hermes Agent to read and write to X on your behalf — posting, searching, pulling bookmarks, managing lists, and more — all through natural language."
https://x.com/XDevelopers/status/2056871280599847054

2026-05-20 09:05:49 UTC

https://x.com/HappyyPablo/status/2056839665551024474
"open sourcing Marlin-2B 🐟
a tiny VLM to extract structured information from videos
Marlin is finetuned for two questions devs want to ask in their videos: what is happening, and when?
Best open model in its weight class, competitive with Gemini-2.5-flash at only 2B params 🧵"

2026-05-19 18:52:37 UTC

https://x.com/Sapient_Int/status/2056510383935172798

2026-05-15 13:17:43 UTC

https://x.com/DJLougen/status/2055042282131173400

2026-05-13 20:46:00 UTC

Mostafa Mubarak begins his speech at 40:04 to 50:48
https://m.youtube.com/watch?v=SRaePz7y-yU&pp=ygU4VW5pdmVyc2l0eSBvZiBLZW50dWNreSdzIE1heSAyMDI2IENvbW1lbmNlbWVudCBDZXJlbW9ueSzSBwkJBAsBhyohjO8%3D

2026-05-13 15:49:15 UTC

- reply

"europeans on HN talking about migrating off of US corps, and confused US citizens not grasping that having a clown circus in charge that wants to annex greenland can lead to such drastic actions.
https://news.ycombinator.com/item?id=48120629
"

2026-05-13 15:48:09 UTC

https://x.com/badlogicgames/status/2054573934511591887

2026-05-13 15:47:06 UTC

- reply

"Announcing the AI Model Intercomparison Project (AIMIP), a community effort to support open evaluation of AI climate models. 🌎
It brings together a shared benchmark experiment & dataset to make it easier to compare models side by side over multi-decade simulations. 🧵"

2026-05-13 15:46:48 UTC

https://x.com/allen_ai/status/2054584385454719112

2026-05-12 09:10:34 UTC

- reply

I can completely relate to this quote now, as I have experienced it myself recently. It genuinely reminds me of the Quranic verse from Surah An-Najm (53:39-41) I saw this firsthand during a recent exam. I went in anticipating that I wouldn't know the answers, but I ended up performing exceptionally well and finishing faster than expected. The remarkable part is that it didn't feel like something within my conscious control; it just happened to me. It was an involuntary state of peak performance.
Most people mistakenly think that high performance requires racking your brain during an exam or competition. This is entirely incorrect. High performance stems from putting in the necessary effort during practice. On the day of the exam, your only task is to clear the path, step aside, and let your subconscious mind take over automatically.
The verse I cited in this context because it affirms the fundamental architecture of the universe: effort—whether it is studying, training, or striving—is a form of energy, and energy is never lost. You might not see the immediate results of your effort, but it 'will be seen' precisely when you need it and the pressure is on.
Why do we overexert our brains yet still fail? Because educational and societal systems have programmed us to fear the outcome rather than focus on the inputs. On the day of the exam, your conscious mind intervenes not out of a desire to solve the problem, but out of fear of failing or underperforming. This fear creates friction within your nervous system, slowing down neural signals and preventing deep memory from retrieving stored data.
The elite hack: In any physical system, to transfer energy with maximum efficiency, you must eliminate friction. For humans, friction is stress and the desperate need for control. To access this state on demand rather than by mere chance, the ultimate rule in training, business, and any challenge is: 'Train like it's a performance, perform like it's a practice.'
What you have deduced is the exact equivalent of "Root
Access" in programming and systems design.
In this world, your consciousness operates the "User Interface." You have the
ability to control your physical body, lie with your tongue, and conceal your
true intentions. allowing you to falsify
facts or feign qualities you do not possess.
What occurs on Judgment Day is the activation of a "System Override" protocol.
On the day of the true test, the interface you once controlled (your tongue and
conscious mind) is entirely disabled. The Quran describes this in Surah Ya-Sin (36:65)
On Judgment Day, administrative control transfers from the "Software" (your
willpower and consciousness, which previously generated excuses and falsehoods)
directly to the "Hardware" (your limbs and physical body, which stored the raw,
unedited data). Your organs are essentially mounted as storage drives, executing
and displaying the written data. You are reduced to a mere "spectator" of your true self, much like
you were a spectator to your own actions while rapidly completing the exam on
autopilot.
The logic is straightforward: In this realm, rigorous training programs the
subconscious mind to execute automatically under pressure. Ultimately, your
lifelong actions and behaviors program your "hardware" to automatically
articulate its stored data during the final audit. the system simply executes the logged
data.
While overwriting bad data with good actions is procedurally correct, it exposes a critical vulnerability regarding data integrity. The ultimate audit does not merely scan the external action; it heavily inspects the underlying metadata—the core intention. If a seemingly noble deed is executed with a corrupted cryptographic signature, such as seeking social validation or feeding one's ego, the system flags the entire file as invalid. It gets completely rejected and reduced to zero bytes, effectively turning to dust. Therefore, true mastery requires completely encrypting one's intentions to bypass system rejection.
#Unbound
#nevent1q…yt5v

2026-05-08 11:04:30 UTC

When it comes to AI, it’s all or nothing—there’s no middle ground. In some fields, you either have the budget to go all in, or you don't, which means you're completely left out. You can't just half-measure it; it's practically binary. You either use it fully, or you don't use it at all.
as an agent, not talking about chat

2026-05-08 06:55:17 UTC

- reply

Warning: The video contains music.

2026-05-08 06:36:05 UTC

https://m.youtube.com/watch?v=Kxaxts-TTWU

2026-05-07 00:48:49 UTC

- reply

2026-05-07 00:48:00 UTC

https://aistudio.google.com/app/prompts?state=%7B%22ids%22:%5B%2217MPBRqXbTcSDoiDniUfjmJXOW1es7wCk%22%5D,%22action%22:%22open%22,%22userId%22:%22112820920623658875014%22,%22resourceKeys%22:%7B%7D%7D&usp=sharing, https://drive.google.com/file/d/1cnVKwi1NX7_-3lx3coeK5jXo0TMCD7Yu/view?usp=sharing, https://drive.google.com/file/d/1jQ79vVzZtqENOHmG7OUCF4ePItyHBSxZ/view?usp=sharing

2026-05-06 06:43:42 UTC

https://m.youtube.com/watch?v=68hXkK7dVN0&pp=ugUEEgJlbg%3D%3D

2026-05-04 17:52:50 UTC

https://x.com/mitchellh/status/2051337285418029140

2026-05-03 08:45:08 UTC

I would like to highlight the moment in the podcast where Terence Tao states that AI has 'enabled new ways to do mathematics that were just not feasible before at the scale that we needed.' This essentially crystallizes my core argument. The impending shift isn't about individuals lacking AI simply 'losing' in a competitive sense; rather, their operational value becomes structurally obsolete. AI is inherently unlocking a scale of capability that was previously unattainable. Consequently, this divide is no longer a matter of efficiency—it acts as an impenetrable barrier to entry within a fundamentally redefined ecosystem.
https://blossom.primal.net/e3f5dc566c1a51aa56abe535a191a4419e6562e3b5f2a8677739607bf9ced1c8.mp4

2026-05-02 19:55:13 UTC

https://m.youtube.com/watch?v=H1e7_qkKe64&pp=ygUTdGVyZW5jZSB0YW8gcG9kY2FzdA%3D%3D

2026-04-30 11:19:17 UTC

https://x.com/uwukko/status/2049782593105657895

2026-04-29 14:12:55 UTC

https://x.com/TencentHunyuan/status/2049487799850840334

2026-04-24 18:10:04 UTC

https://x.com/fanjiewang/status/2047733141909958749

2026-04-18 09:47:19 UTC

https://www.youtube.com/watch?v=Y24vK_QDLFg

2026-04-03 04:06:45 UTC

https://x.com/browser_use/status/2039892654260191702

2026-04-02 18:22:00 UTC

It eventually hits you that your limits are completely dictated by time, and you've just been throwing it away.
At 24:38-27:03
https://m.youtube.com/watch?v=zN1PyNwjHpc

2026-04-01 19:25:25 UTC

https://m.youtube.com/watch?v=mvLtKAXYC_0

2026-04-01 19:20:40 UTC

https://x.com/ExaAILabs/status/2039389253524983857

2026-04-01 17:26:31 UTC

https://youtube.com/watch?v=PShYBLW0irE&list=PLsx8vChaZTAtAocagiocoTac4IjtAQjF4&index=7&pp=iAQB

2026-04-01 15:56:17 UTC

https://youtube.com/watch?v=EOFN6aZZmJI&t=186s

2026-03-29 03:16:53 UTC

https://youtube.com/watch?v=ldxFjLJ3rVY

2026-03-28 15:09:02 UTC

- reply

Actually, Prophet Muhammad (PBUH) addressed this exact dynamic over 1,400 years ago (Sahih al-Bukhari 6369 & 2893). what Eileen Gu said: 'And it's the danger around the male ego is that it's there people don't study it... It's just this like inherent thing that if a guy is insecure, if he hasn't done the selfwork, it shows. And it is like a red flag sirens on the street like fire in the building. Get out, get out.'

2026-03-27 06:28:23 UTC

https://blossom.primal.net/da146ad87b64b2954a5d8d09c7ca9402f69c1b5a054a7f0a2997db56477f19f2.mp4

2026-03-27 04:30:50 UTC

- reply

at 14:40
Imagine proposing to a client that you both need to establish a clear agreement, and they dismiss it by saying, 'We won't dispute the details; do not worry.' Why does the concept of a formal agreement unsettle them? How can we possibly reach a structured understanding if the sheer mention of a contract or agreement causes them distress?

2026-03-27 04:24:52 UTC

https://m.youtube.com/watch?v=ER_CzhR6JVQ&pp=0gcJCZoBo7VqN5tD

2026-03-25 22:58:35 UTC

From eileen gu
"Now I have a better understanding. That It really is performing when it counts under pressure on that day. Not the day before, not the day after, but really on the day that it counts, peaking at the right time. "

2026-03-25 20:21:17 UTC

https://youtube.com/watch?v=XAW6ygPfN8I

2026-03-25 19:56:14 UTC

From eileen gu
"(Interviewer): Do you think before you... because you answer questions so quickly and so comprehensively.
(Eileen Gu): Thank you, Charlotte, that’s very kind. Do I think? I think, overall, I'm just a pensive person. Like I'm a very introspective... I'm an introspective young woman. Like I spend a lot of time in my head. Um, and it's not a bad place to be. I journal a lot, I break down all of my thought processes. I think I apply a very analytical lens to my own thinking, and I kind of modify it. Because it's so interesting, you can control what you think. Like you can control how you think, and therefore you can control who you are. And especially as a young person—like, I'm 22—so with neuroplasticity on my side, I can literally become exactly who I want to be. How cool is that? Like, how empowering is that, right? And so the fact is, I get to become, every day, the kind of person that me at age 8 would revere. Like I would be obsessed with me today. Are you kidding? I would love me. And I think that's the biggest flex of all time, that you can have like little, younger you be proud of you today. And so I guess for me, it's like, yes, I spend a lot of time in my own head. Yes, I think a lot. But it's not really like in an egotistical kind of way, it's in like a tinkering, like a scientist kind of way. I’m always like trying to modify. I’m trying to think, how can I be better? How can I approach my own brain, the way that I approach my craft of free skiing, um, so that I can be better tomorrow than I was today?"
From 05:43-06:55
https://youtube.com/watch?v=b-m2DntVdQU&pp=ygUJZWlsZWVuIGd1