ehab

tring nostr

Public Key

npub10vap6lgajn4j22rpgrq686vsktzmes7jkj7mavwy4yrdfz3hpsnqxafruw

Profile Code

nprofile1qqs8kwsa05wef6e99ps5psdraxgt93duc0ftf0d7k8z2jpk53gmscfspzemhxue69uhhyetvv9ujuurjd9kkzmpwdejhgqg5waehxw309ahx7um5wghryc3ewsh8s7t6tjkfpn

Publishing to

relay.primal.net

nostr.2b9t.xyz

lnbits.papersats.io/nostrclient/api/v1/relay

relayable.org

relay.cosmicbolt.net

nostr-verif.slothy.win

Author Public Key

npub10vap6lgajn4j22rpgrq686vsktzmes7jkj7mavwy4yrdfz3hpsnqxafruw

Show more details

Last Notes

2026-07-22 18:10:27 CEST

https://x.com/sudoingX/status/2076607813485564275
"i think framework desktops are so underrated for what they cost and what they can do. this strix halo flagship, 128gb unified, was sent to me by framework and i've been waiting for the failure point ever since. still waiting. it's zero.
it runs huge moe models at genuinely good speed, quiet enough that i forget it's on, compact enough that it disappears on the desk, and it's now the main driver between my two monitors. dgx spark stays the ultimate box for training and clustering, connectx and scale is nvidia's world, but for personal starter inference this thing feels unbeatable at the price.
and the part i did not expect, it games. proper steam games running on vulkan, i plugged in two controllers and lost an evening to it.
one quiet box that serves a 397b moe at 18 tok/s, holds a 4 million token context on a 35b, and runs your steam library. the local ai box i keep telling people to wait for kind of already exists."

2026-07-22 18:05:32 CEST

https://x.com/Teknium/status/2079760273641185662

2026-07-22 18:00:11 CEST

https://m.youtube.com/watch?v=434cG4g5KLE&pp=0gcJCWQCo7VqN5tD

2026-07-22 17:52:15 CEST

https://frame.work/desktop?tab=192gb-coming-soon

2026-07-22 09:03:25 CEST

- reply

I'm using it on "Q8_0" quantization, since I have only 12GB vram

2026-07-22 08:59:22 CEST

My model " Ornith-1.0 9B Dense models"
the best model I used on my RTX 3060 12GB
https://deep-reinforce.com/ornith_1_0.html

2026-07-22 08:40:57 CEST

https://x.com/AMDRyzen/status/2079713033987784860
"HotHardware
reviews AMD Ryzen AI Halo, citing excellent build quality, software flexibility and competitive performance in a compact 128GB local AI workstation that “easily earns” its Recommended award.
Dive into the review
https://hothardware.com/reviews/amd-ryzen-ai-halo-hands-on-review?page=4
"

2026-07-21 20:37:02 CEST

https://x.com/dosco/status/2079292484660969902
https://x.com/a1zhang/status/2079203524395573442
"DSPy, RLM, PEEK, GEPA is context engineering, you can compose, measure, eval and improve!
keep in mind you have to divide a task so every piece lands in distribution aka small familiar canonical calls. weird bespoke sub-prompts don't generalize no matter how small they are.
this is where DSPy signatures and optimizers like GEPA come in, they convert an infinite space in a handful of prompt shapes that's the equivalence class the model sees a template with different content.
an optimizer cannot just search all possible prompts, signatures are what make the prompt space searchable at-all.
when a composed program underperforms you can easily eval each component against its own metric to find the one that's off distribution."
"Transformers struggle to generalize to tasks they were not explicitly trained on. Instead, we propose in 2026 that it is the job of the harness to generalize through composition.
We observe a powerful property when training RLMs: for tasks with shared structure that look different, the root model naturally learns the same trajectory, meaning it views the two task trajectories as the same! In other words, the Transformer does not need additional generalization capabilities to transfer capabilities from one task to the other, the harness induces it.
We find that well-designed harnesses form a quotient set over task trajectories, meaning their individual LLM calls can see structurally “similar” tasks as near-identical, token-for-token! Harnesses can effectively generalize for the Transformer during training, without relying on any intrinsic generalization capability from the model.
For example, RLMs can see problems of different lengths as the same: we show that RLMs can train exclusively on short tasks, and fully generalize to similar but unseen tasks 8-32x longer because it produces near identical trajectories for both.
Taking this further, we show that tasks across different domains (e.g. math solutions vs. essay writing) that share a decomposition strategy exhibit the same generalization effect. RLMs can train on the problem of finding which essays belong to the same author and improve performance on finding math problems that share similar solutions.
The full blogpost, experiments, and discussion are in the thread below."

2026-07-21 20:19:10 CEST

so true
https://x.com/karpathy/status/2079610838143623371
"One pattern I find useful for working with LLMs is a nice long ramble session. Sometimes the LLM needs more bits to understand what you're trying to achieve, but you're too lazy to type them. In these cases I like to lean back, switch to /voice and just ramble for like 10 minutes, total mess, anything goes, full stream of consciousness. Sometimes I declare it up top, something like "switching to speech recognition sorry for any typos...". Sometimes I turn it into a small interview of a few turns. But I find that the LLMs are somehow very good at reconstructing long incoherent rambles and often their echo of your own tangle of thoughts comes out quite a bit cleaner than what you started with. The result is that you improve the mind meld and have to correct things less from that point on."

2026-07-21 20:16:25 CEST

https://x.com/thdxr/status/2078620452075733302
"being pro-opensource doesn't mean you have to be anti-proprietary
it's fine for proprietary options to exist, they usually invent the category and make it possible for the opensource option to emerge
the only thing that matters is the opensource option is allowed to exist"

2026-07-21 20:05:28 CEST

https://x.com/maria_rcks/status/2079622096456446211
"let this be a reminder that you can have EVERYTHING and still lose horrendously"

2026-07-21 18:47:29 CEST

https://x.com/uwukko/status/2079596604441444667
"i wonder who seeded the narrative that browser fingerprinting can be "blocked", because that's simply not true if you keep the browser functional"

2026-07-21 17:20:17 CEST

https://x.com/typecraft_dev/status/2078883155025490158

2026-07-21 16:04:39 CEST

https://x.com/jartine/status/2078791718728593453
"One of the things that helped me build cosmopolitan libc was not having a cell phone until 2023. I didn’t even have a drivers license or a passport when I was working with Mozilla and Google. I spent all my time in GNU Emacs and GNU Screen and PuTTY."

2026-07-21 15:58:09 CEST

https://deeplearningwithpython.io/

2026-07-21 12:06:09 CEST

https://x.com/Grantblocmates/status/2079317248058020339
"Monid is a great product, man
If the future truly is headless
Subscriptions will be a thing of the past
Everything is pay per ping"

2026-07-20 23:24:42 CEST

https://x.com/shengkun_ye
https://x.com/shengkun_ye/status/2079289736250970258
"We just killed Apify.
Your agent can now read every social media platform.
No logins. No subscriptions.
X, Reddit, LinkedIn, TikTok, Facebook, Instagram, YouTube, Rednote, and even Amazon.
Apify: $199/month.
Monid: From $0.0015/request. Pay as you go.
"
https://monid.ai/blog/social-media-data

2026-07-20 23:19:27 CEST

https://x.com/ctnzr/status/2079307416970408175

2026-07-20 21:16:24 CEST

the video contains music.
https://blossom.primal.net/8830096df2374b07a9e6dca79b8f62dda90cca2d3e6370172c06ed0502872efe.mp4

2026-07-20 20:20:59 CEST

https://x.com/MikeBradleyAI/status/2079048543356985808
"Open models :
1. Massively increase innovation
2. Decrease costs of experimentation
3. Create a much larger AI economy
4. Democratize intelligence
5. Improve competition
6. Accelerate hardware investments
7. Accelerate scientific progress
8. Reduce bottlenecks"

2026-07-20 20:15:17 CEST

https://x.com/jun_song/status/2079230646925640057
"If you don’t own 256GB of VRAM or unified RAM right now,
You will get into serious trouble soon.
check back in 6 months"

2026-07-20 11:41:20 CEST

- reply

as the ai father said

2026-07-20 11:36:58 CEST

I just need llms, for its knowledge.
i only have ............. secs life

2026-07-19 11:36:48 CEST

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

2026-07-19 07:37:53 CEST

"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

2026-07-19 01:20:01 CEST

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

2026-07-13 20:50:30 CEST

https://x.com/p0/status/2076690147144503390
"Introducing Parallel Search Turbo: the fastest and most affordable way to bring the web to your AI."

2026-07-13 14:13:21 CEST

- reply

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

2026-07-13 14:06:41 CEST

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

2026-07-13 12:49:22 CEST

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

2026-07-12 20:41:56 CEST

https://m.youtube.com/watch?v=VDQTuJWST4M
"Forget Discord, I'm switching back to AIM"

2026-07-11 00:08:39 CEST

https://m.youtube.com/watch?v=Jo4Jpx_N4Ms
"سر البحث العميق على الأنترنت | الاستخبارات مفتوحة المصدر | باختصار"

2026-07-03 15:59:26 CEST

"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 21:42:50 CEST

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

2026-06-24 15:58:58 CEST

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

2026-06-19 11:14:35 CEST

Long term maintenance is the real deal.

2026-06-17 20:27:19 CEST

https://m.youtube.com/watch?v=m_fJ41Ecewc&pp=0gcJCUACo7VqN5tD

2026-06-17 19:33:44 CEST

https://m.youtube.com/watch?v=ML3q7Ok4hJg

2026-06-17 08:38:37 CEST

https://m.youtube.com/watch?v=8ZUm4nuxFfs&pp=ygUNQ2hpbWVyYSBMaW51eA%3D%3D

2026-06-16 17:16:37 CEST

- reply

Here is the architectural and psychological anatomy of this system's future:
1. Architectural Dimension: The Deterministic State Crisis
- Imperative vs. Declarative: Gentoo relies on an imperative model. You
instruct the system step-by-step (via USE flags and emerge). The engineering
flaw here is 'state mutation'. After a year of use, your Gentoo system
becomes a non-reproducible entity.
- The Real Killer (NixOS & Guix): The future is heading towards declarative
systems. In NixOS, you define what you want in a single file, and the system
builds itself. NixOS literally hijacked Gentoo's target audience
(control-freak engineers) because it offers the same deep customization, but
with the advantage of exact reproducibility and rollbacks. Architecturally,
NixOS is the natural evolution that makes Gentoo's philosophy look like
Stone Age technology.
2. Technical Dimension: Supply Chain Security
- The Illusion of Local Build Security: Previously, compiling software from
source on your own machine (as Gentoo does) was considered peak security
because you didn't trust external binaries.
- The 2026 Reality: This concept has collapsed. Today, security relies on
reproducible builds. If your local compiler is compromised, everything you
build is poisoned. Modern systems rely on isolated, cryptographically signed
build servers. Gentoo is trying to catch up by recently offering binary
packages, but this defeats its core purpose.
3. Psychological Dimension: Sunk Cost Fallacy
- A desktop Gentoo user suffers from a massive cognitive load. They spend 20%
of their time maintaining the system, resolving circular dependencies, and
reading eselect updates.
- This creates an 'illusion of achievement'. The user feels smart for fixing a
complex system, whereas a true engineer knows an OS should be invisible to
allow focus on actual work. This psychological burden is what drives
professionals to abandon Gentoo once they enter the real job market.
4. Edge Case Analysis: Where Gentoo Survives If Gentoo is dying on the desktop,
where does it live? Answer: In the shadows.
- Case 1: Embedded Systems & ChromeOS. Google uses Portage (Gentoo's package
manager) to build ChromeOS. Why? Because Portage is brilliant at
cross-compiling (building an ARM system using an x86 server). Here, Gentoo
is not an OS; it is an OS build framework. This is its true future.
- Case 2: RISC-V and Custom Silicon. With the rise of open RISC-V processors,
companies are manufacturing highly customized silicon. There are no
ready-made binaries for these chips. Here, Gentoo steps in. It is the
perfect system to bootstrap a completely new hardware architecture from
scratch.
- Case 3: High-Frequency Trading (HFT). In stock markets, a microsecond means
millions of dollars. Some firms use Gentoo to build a highly customized
Linux kernel, stripping the system of everything except what the trading app
needs, compiled specifically for the server's CPU cache.
The Insider Verdict: Does Gentoo have a future? Yes, but not as a consumer Linux
distro.
- Its future as an infrastructure tool for companies building embedded devices
(IoT, Routers, Chromebooks) is bright and guaranteed.
- Its future as a desktop OS or general server is clinically dead. Technical
debt, power consumption, and the rise of declarative systems (NixOS) and
immutable systems (like Fedora Silverblue) make Gentoo on the desktop
architecturally unjustified.
The Radical Solution: If you are considering Gentoo for 'absolute control',
stop. Go learn NixOS. You will write your entire system as code (Infrastructure
as Code), gain deeper control than Gentoo, and your system will never break.
However, if you want to build an OS for a router manufactured by your company,
Gentoo (Portage) is your ultimate weapon.
However, to escalate: NixOS is not the endgame. The Nix language is flawed, overly complex, and prone to spaghetti code.
The deeper solution is Guix. It utilizes Scheme (Lisp), which is a fully-fledged programming language, not merely an advanced JSON.
The true future of the desktop lies in an immutable base combined with containers.
Never install tools directly on the host system.
Utilize Fedora Silverblue or Alpine.
Development environments should be confined within Distrobox or Devcontainers.
The host system functions solely as a hypervisor.
This ensures total isolation, zero dependency conflicts, and zero crashes. The system remains perpetually operational.

2026-06-16 12:26:56 CEST

"i hope it's clear now why open source models are important
i've said before i can respect the position around safety but it's completely naive
even if you think you have superior morality and should control it someone will kick you out and take control"
https://x.com/thdxr/status/2066764365819261083

2026-06-16 11:10:36 CEST

I'm on Chimera Linux now

2026-06-16 10:55:36 CEST

https://m.youtube.com/watch?v=8HBDE-msUjw

2026-06-15 21:57:57 CEST

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

2026-06-15 21:48:25 CEST

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

2026-06-15 15:07:49 CEST

https://m.youtube.com/watch?v=7M00ZDCjAA8&pp=ygUNQ2hpbWVyYSBMaW51eA%3D%3D
"The Linux & BSD Distro Tier List (20XX Edition)"

2026-06-13 21:46:40 CEST

- reply

https://x.com/browser_use/status/2065874336238776745

2026-06-13 21:46:05 CEST

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 19:30:45 CEST

- reply

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

2026-06-13 16:32:21 CEST

- reply

Mistakes are inevitable. You rarely get it right on the first try. In fact, trial and error is exactly how we learn and remember things.
Just re-reading and repeating stuff is a total waste of time. Pros actually use Active Recall and Spaced Repetition. Here's how it works: quiz yourself -> mess up at first -> your brain catches it and fixes it -> it locks into your long-term memory. You can't really learn without a bit of struggle and making mistakes.

2026-06-13 16:18:40 CEST

- reply

I need chat for long context = things that I really don't know, and need mentor

2026-06-13 14:25:25 CEST

https://m.youtube.com/watch?v=Nzmj7K0FNRY&pp=ugUEEgJlbg%3D%3D

2026-06-13 09:48:57 CEST

ai agents free
ai search engines free
ai chat is partly free

2026-06-13 03:21:43 CEST

"AMD Ryzen AI Halo. The ultimate local AI developer platform.
Pre-order now: https://bit.ly/4xv5PJS
⚡ Up to 128GB unified memory
⚡ Support for models up to 200B parameters
⚡ Windows & Linux support
⚡ Ready-to-run AI workflows out of the box
Build, prototype, and deploy locally without cloud constraints."
https://x.com/AMD/status/2065479151969050856

2026-06-12 19:36:17 CEST

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

2026-06-12 18:59:18 CEST

"AMD's Strix Successor Just Caught the M4 Pro"
https://m.youtube.com/watch?v=sxMSKyrnZH4

2026-06-12 18:29:01 CEST

"Scira TUI is here!
❯ http://cli.scira.ai"
https://x.com/zaidmukaddam/status/2065454339871961110

2026-06-12 15:52:57 CEST

https://x.com/thdxr/status/2065099763985293742
"it's not done if it's not implemented
it's not done if the implementation is ugly
it's not done if it's not documented
it's not done if users can't discover it
it's not done if you can't market it"

2026-06-12 12:57:50 CEST

- 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.

2026-06-12 12:57:01 CEST

- reply

devs know it's totally fine to not memorize a massive million-line codebase. You just work on massive code by knowing the parts you touch, without holding the whole thing in your head.

2026-06-12 11:18:41 CEST

https://m.youtube.com/watch?v=WnBx1Vi7M6w

2026-06-12 10:28:30 CEST

Gentoo is exactly what Arch promised to be.

2026-06-12 09:50:06 CEST

https://m.youtube.com/watch?v=E0spRyDu5kk&t=1s

2026-06-12 09:20:35 CEST

- reply

i need to learn how I memorize things

2026-06-12 08:56:00 CEST

https://m.youtube.com/watch?v=E0spRyDu5kk&t=1s

2026-06-11 17:21:20 CEST

Gentoo, the end of traditional Linux distros.

2026-06-11 12:43:11 CEST

"Got a lot of DMs and comments asking how to run
@liquidai
LFM2.5-8B-A1B with Hermes Agent locally, so I put together a complete step bystep guide. It covers macOS, Linux, and Windows. Hope it helps everyone who reached out."
https://x.com/noctus91/status/2064762556225556837

2026-06-10 10:32:53 CEST

"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

2026-06-10 07:12:18 CEST

"Best models for your hardware this week.
8-12GB
- https://huggingface.co/LiquidAI/LFM2.5-8B-A1B incredible model, so fast, so small
16-32GB
- latest Google model, Gemma 12B: https://huggingface.co/google/gemma-4-12B really solid performance up neck and neck with a model 2x its size from a month ago.
Jetbrains new model, best in class on livecode bench
32-96gb
- Nex-N2-Mini GPT style postrain of Qwen-35B it seems to be its class leader caveman style reasoning https://huggingface.co/nex-agi/Nex-N2-mini
- Jackrong’s Qwopus is the #1 overall Q4 of Qwen3.6-27B on our benchmark suite of 5 agent + coding benchmarks (1200 samples total) https://huggingface.co/Jackrong/Qwopus3.6-27B-v2-MTP-GGUF
192gb
- Step-3.7-Flash is hard to beat, high scores, really fast inference, vision capable, later cutoff dates https://huggingface.co/stepfun-ai/Step-3.7-Flash
384gb
- Nex-N2-Pro GPT style post train of Qwen-3.5-397B incredibly strong and #1 on deepswe if their claims are right https://huggingface.co/nex-agi/Nex-N2-Pro
768gb
- very promising post-train of GLM-5.1 that wins out on 8 benchmarks https://huggingface.co/mindlab-research/Macaron-V1-Preview-749B
"
https://x.com/0xSero/status/2064220082264293820

2026-06-09 21:51:05 CEST

https://m.youtube.com/watch?v=0uwRFSGFJgM&pp=ugUEEgJlbg%3D%3D
"Congratulations, you’ve reached the end of the Internet!"

2026-06-09 16:57:48 CEST

It is necessary to learn methods of memorization.
I really need to figure out how to memorize things better.

2026-06-08 22:03:53 CEST

- reply

https://blossom.primal.net/a8c68dc69ab6459fabd939a2c3148b610b5f117d5fc75d1594f72b1b3c3d049a.jpg

2026-06-08 22:03:38 CEST

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

2026-06-08 10:48:40 CEST

https://m.youtube.com/watch?v=fi29pfLcW4I

2026-06-08 10:34:58 CEST

https://m.youtube.com/watch?v=4_x0USuQ9Ss&pp=ugUHEgVlbi1VUw%3D%3D

2026-06-08 10:18:54 CEST

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 21:58:11 CEST

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 21:38:02 CEST

- 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.

2026-06-07 20:10:36 CEST

- reply

That’s exactly why I switched from NixOS to Gentoo, btw

2026-06-07 20:02:42 CEST

https://m.youtube.com/watch?v=EnP1rdx-5So&pp=ugUEEgJlbtIHCQlAAqO1ajebQw%3D%3D
"Linux Is Becoming Android (And Almost No One Noticed) | Source Code Ep. 21"

2026-06-06 23:10:43 CEST

"Introducing Harness-1, a 20B search agent trained with a state-externalizing harness.
> frontier-level long-horizon search, rivaling Opus-4.6 and outperforming GPT-5.4
> Context-1-level cost and latency
> externalizes candidates, evidence, verification, and search history
> open-source"
https://x.com/patpcj/status/2063298457398636570

2026-06-06 17:40:12 CEST

- reply

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

2026-06-06 17:40:00 CEST

- 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

2026-06-06 17:39:37 CEST

- reply

https://blossom.primal.net/3fc1668df43c9cd7aa2a0be016b89df7795d2c445c85541a8e5004d42d6d6d39.jpg
https://blossom.primal.net/0ffdf43dc7c798c861b950e2342f78686bd1042d1b63b20c81a604cc386159dd.jpg

2026-06-06 13:18:57 CEST

https://x.com/mudler_it/status/2063207757692441056

2026-06-06 11:49:53 CEST

https://m.youtube.com/watch?v=_sQLB3fZn7I
"اعتياد النعم والألم والعبادات وملل الزواج"

2026-06-06 08:01:10 CEST

https://m.youtube.com/watch?v=kJUIO2sLWlw
"تجارب صغيرة لتحسين حياتك - علي وكتاب"

2026-06-05 22:19:10 CEST

- reply

أكثر من 20 دراسة علمية حقيقية (منشورة في مجلات علمية محكمة مثل PubMed، PMC، Sleep، وغيرها) تم الاعتماد عليها في إعداد الفيديو.**
المواضيع الرئيسية : علم الأحياء الزمني (Chronobiology)، الدورات فوق اليومية (Ultradian Rhythms) ودورة الراحة-النشاط الأساسية (BRAC) لناثانييل كلايتمان، إدارة الطاقة مقابل إدارة الوقت، ذروة الكورتيزول الصباحية وتأثير الضوء الصباحي، الانخفاض الفسيولوجي بعد الغداء (Postprandial Dip)، قاعدة 20-20-20 لإجهاد العيون، والدوبامين الناتج عن إنجاز المهام.
اخترت دراسات peer-reviewed (معظمها من PubMed/PMC) مع روابط مباشرة حقيقية وأحدث ما يمكن (حتى 2024-2025). الدراسات تؤكد أن الإنتاجية مرتبطة بالإيقاعات البيولوجية أكثر من "تقسيم الوقت إلى مربعات"، وأن احترام الدورات الطبيعية (90-120 دقيقة) + التعرض للضوء الصباحي + الراحة يحسن التركيز والطاقة.
1. الدورات فوق اليومية (Ultradian Rhythms) وBRAC (كلايتمان)
Kleitman, N. (1982). Basic rest-activity cycle--22 years later. Sleep, 5(4), 311-317.
https://pubmed.ncbi.nlm.nih.gov/6819628/
(الدراسة الأساسية التي أعاد فيها كلايتمان تأكيد دورة BRAC الـ90-120 دقيقة في اليقظة والنوم).
Neubauer, A.C. (1995). Ultradian rhythms in cognitive performance: no evidence for a 90-minute cycle. Biological Psychology, 40(3), 259-272.
https://pubmed.ncbi.nlm.nih.gov/7669837/
(دراسة تجريبية على 60 شخصاً اختبرت الدورات في الأداء المعرفي).
Okudaira, N. et al. (1984). No basic rest-activity cycle in head, wrist or ankle. Physiology & Behavior, 32(5), 843-846.
https://pubmed.ncbi.nlm.nih.gov/6494289/
(اختبار BRAC في الحركة اليومية).
2. علم الأحياء الزمني (Chronobiology) وإدارة الطاقة مقابل الوقت
Yan, R. et al. (2024). Identifying Links Between Productivity and Biobehavioral Rhythms Modeled from Passively Collected Mobile Data Streams. PMC.
https://pmc.ncbi.nlm.nih.gov/articles/PMC11066747/
(ربط الإنتاجية بالإيقاعات البيولوجية عبر بيانات الهاتف).
Summa, K.C. & Turek, F.W. (2014). Interactions between Circadian Rhythms and Energy Regulation. Advances in Nutrition.
https://pubmed.ncbi.nlm.nih.gov/24829483/
(تفاعل الإيقاعات اليومية مع الطاقة والتمثيل الغذائي).
Gonnissen, H.K.J. et al. (2013). Chronobiology, endocrinology, and energy. Obesity Reviews.
https://pubmed.ncbi.nlm.nih.gov/23387351/
(دور الإيقاعات في توازن الطاقة).
Volk, S. et al. (2023). Circadian leadership: A review and integration of chronobiology and leadership. Journal of Organizational Behavior.
https://onlinelibrary.wiley.com/doi/full/10.1002/job.2659
(تطبيق الـChronobiology على الإنتاجية والقيادة).
Reddy, S. et al. (2023). Physiology, Circadian Rhythm. StatPearls - NCBI Bookshelf.
https://www.ncbi.nlm.nih.gov/books/NBK519507/
(مراجعة شاملة للإيقاعات اليومية وتأثيرها على الطاقة).
3. الكورتيزول الصباحي والتعرض للضوء (Morning Cortisol + Sunlight)
Jung, C.M. et al. (2010). Acute Effects of Bright Light Exposure on Cortisol Levels. Journal of Biological Rhythms.
https://pmc.ncbi.nlm.nih.gov/articles/PMC3686562/
(تأثير الضوء الساطع على الكورتيزول).
Scheer, F.A. et al. (1999). Light affects morning salivary cortisol in humans. Journal of Clinical Endocrinology & Metabolism.
https://pubmed.ncbi.nlm.nih.gov/10487717/
Leproult, R. et al. (2001). Transition from dim to bright light in the morning induces an immediate elevation of cortisol levels. Journal of Clinical Endocrinology & Metabolism.
https://pubmed.ncbi.nlm.nih.gov/11231993/
Figueiro, M.G. et al. (2012). Short-Wavelength Light Enhances Cortisol Awakening Response. PMC.
https://pmc.ncbi.nlm.nih.gov/articles/PMC3413982/
Bowles, N.P. et al. (2022). The circadian system modulates the cortisol awakening response. PMC.
https://pmc.ncbi.nlm.nih.gov/articles/PMC9669756/
Petrowski, K. et al. (2019). The effects of post-awakening light exposure on the cortisol awakening response. Psychoneuroendocrinology.
https://pubmed.ncbi.nlm.nih.gov/31220690/
4. الانخفاض بعد الغداء (Postprandial Dip / Afternoon Slump)
Monk, T.H. (2005). The post-lunch dip in performance. Chronobiology International.
https://pubmed.ncbi.nlm.nih.gov/15892914/
(دراسة كلاسيكية تثبت أنها ظاهرة بيولوجية حتى بدون غداء).
Monk, T.H. (1996). Circadian determinants of the postlunch dip in performance. Journal of Biological Rhythms.
https://pubmed.ncbi.nlm.nih.gov/8877121/
Romdhani, M. et al. (2018). Sleep deprivation affects post-lunch dip performances. PMC.
https://pmc.ncbi.nlm.nih.gov/articles/PMC6413570/
Zhou, Y. et al. (2021). Does Bright Light Counteract the Post-lunch Dip in Alertness? Frontiers in Public Health.
https://www.frontiersin.org/journals/public-health/articles/10.3389/fpubh.2021.652849/full
5. قاعدة 20-20-20 وإجهاد العيون
Talens-Estarelles, C. et al. (2023). The effects of breaks on digital eye strain, dry eye and binocular vision: Testing the 20-20-20 rule. Contact Lens and Anterior Eye.
https://www.sciencedirect.com/science/article/pii/S1367048422001990
Datta, S. et al. (2023). The 20/20/20 rule: Practicing pattern and associations with asthenopic symptoms. PMC.
https://pmc.ncbi.nlm.nih.gov/articles/PMC10391416/
Kaur, K. et al. (2022). Digital Eye Strain- A Comprehensive Review. PMC.
https://pmc.ncbi.nlm.nih.gov/articles/PMC9434525/
6. الدوبامين وإنجاز المهام (Dopamine from Checklists / Task Completion)
Bromberg-Martin, E.S. et al. (2010). Dopamine in motivational control: rewarding, aversive, and alerting. Neuron.
https://pmc.ncbi.nlm.nih.gov/articles/PMC3032992/
Westbrook, A. et al. (2016). Dopamine does double duty in motivating cognitive effort. PMC.
https://pmc.ncbi.nlm.nih.gov/articles/PMC4759499/
Walton, M.E. et al. (2019). What Is the Relationship between Dopamine and Effort? Neuron.
https://pmc.ncbi.nlm.nih.gov/articles/PMC6352317/
Boehler, C.N. et al. (2011). Task-Load-Dependent Activation of Dopaminergic Midbrain Areas. PMC.
https://pmc.ncbi.nlm.nih.gov/articles/PMC6622993/
ملاحظات إضافية:
معظم هذه الدراسات تؤكد أن "إدارة الوقت" التقليدية (مثل time-blocking عشوائي) تتعارض مع البيولوجيا، بينما الـ"energy management" عبر الدورات الـ90-120 دقيقة + الضوء الصباحي + الراحة بعد الغداء + الاستراحات القصيرة (مثل 20-20-20) تحسن الإنتاجية فعلاً.
الدراسات من مصادر موثوقة (PubMed/PMC) ويمكنك تحميلها مجاناً في معظم الحالات.
الفيديو يعتمد على هذه الأسس العلمية (مثل BRAC لكلايتمان والكورتيزول الصباحي)، لكن بعض التطبيقات (مثل Bitrix24) تجارية وليست علمية.

2026-06-05 18:07:31 CEST

https://m.youtube.com/watch?v=RQj3Qlcwacs&pp=ugUEEgJhcg%3D%3D
"إدارة الوقت خدعة! هذا النظام العلمي سيغير حياتك (دليلك للعمل العميق!)"

2026-06-05 13:30:24 CEST

https://m.youtube.com/watch?v=ygscsZ09zPE#
"PewDiePie did it again"

2026-06-05 11:25:21 CEST

"
Second big release from us today: Nemotron-3.5-ASR-Streaming!
🌎40 languages
⚡️80ms - 1s controllable latency
🔥240 - 2400 concurrent streams on 1xH100
🧱FastConformer Cache-Aware RNN-T architecture
https://huggingface.co/nvidia/nemotron-3.5-asr-streaming-0.6b
https://x.com/PiotrZelasko/status/2062538923776290909
"

2026-06-03 06:40:10 CEST

https://m.youtube.com/watch?v=IexMvZt3m5g&pp=ygUGZ2VudG9v
"Gentoo Linux vs Arch Linux vs NixOS (2026)"

2026-06-02 06:11:31 CEST

- reply

I'm back to vllm

2026-06-01 20:59:40 CEST

https://x.com/badlogicgames/status/2061400604103131561
"recommended viewing. andrew kelley, creator of zig, on why he built zig, ai ban, move to codeberg. i like his ethos.
https://m.youtube.com/watch?v=iqddnwKF8HQ&pp=2AEAkAIB
"

2026-06-01 20:07:58 CEST

https://x.com/AMDRyzen/status/2060073899841216842
"AMD Ryzen AI Max PRO 400 Series processors are coming to Ryzen AI Halo platforms in Q3.
Built to support a new class of local AI workloads, Ryzen AI Halo is designed for more agents running in parallel, more intelligence operating locally, and systems that reason and adapt in real time."

2026-05-31 19:53:08 CEST

- reply

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

2026-05-31 18:44:04 CEST

- reply

The Danger of Paralysis by Analysis: Every time you focus on grammar while
speaking (debating whether to use "had been" or "was"), you disrupt your organic
flow. The fear of making mistakes acts as an overprotective algorithm that
forces the system into a deadlock, halting outbound data.
Failure to Comprehend Idioms & Nuances: Literal translation utterly fails to
process idiomatic expressions, sarcasm, and tonal shifts. This data is not
stored in an explicit dictionary; it exists strictly within the emotional
context of the situation.
OS Level Override: There is no such thing as "English study time." Biological
adaptation demands total immersion as a survival primitive. Devices,
smartphones, keyboards, terminals, and technical references—everything in your
environment must be forcefully compelled to operate in English without
exception.
The momentary friction you experience initially is literally the physical pain
of forging new neural pathways (Neuroplasticity). Embrace this friction as a
strict indicator that the system is executing correctly.

2026-05-31 18:33:43 CEST

Honestly, the core concepts are rarely that hard to grasp. The real bottleneck is the massive volume of memory work—and all the redundant noise that comes with it. Your brain biologically requires downtime to actually wire that information together. That biological limit completely explains why cramming the night before is completely useless

2026-05-31 17:10:44 CEST

https://m.youtube.com/watch?v=ShGT-fY7S98&pp=ugUEEgJlbg%3D%3D
"Escaping The Permanent Underclass"

2026-05-31 01:58:23 CEST

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"