Mohammed Momen

Slow but Steady.
Machine Learner.

My scientific interests range from AI and ML to software engineering, Math, and philosophy.

[Mohammed Momen @momen_ai]

Nothing in life is more rewarding than working on the right things with the right people, in the right place, at the right time, for the right purpose.

#momen

[MENA Visuals @menavisualss]

🇾🇪 Yemen, far from the modern world, close to the sky.

Video by Yousef Al-Jabal

[Gergely Orosz @GergelyOrosz]

How have the fundamentals of building large, distributed software systems changed the last decade? A conversation with Martin Kleppmann (author of Designing Data-Intensive Applications) - given that the second, updated edition of the book was just released.

Timestamps:

00:00 Early career
05:46 Building Rapportive
10:47 Working at LinkedIn
14:09 Writing Designing Data-Intensive Applications
23:00 Reliability, scalability, and repeatability
26:24 DDIA: the second edition
30:50 Tradeoffs of using cloud services
39:02 How the cloud changed scaling
42:53 The trouble with distributed systems
49:02 Ethics for software engineers
52:45 Formal verification
1:00:12 Academia vs. industry
1:03:50 Local-first software
1:09:50 Computer science education
1:18:32 Martin’s current research and advice

Brought to you by:

• @statsig  – ⁠ The unified platform for flags, analytics, experiments, and more. https://t.co/ZCSOIcWv31

• @SonarSource – The makers of SonarQube, the industry standard for code verification and automated code review. Check out Sonar's new architecture management capabilities that ensure both humans and AI agents respect your system’s blueprint. https://t.co/3Rycx5lfXE

• @WorkOS – Ship enterprise features – SSO, directory sync, RBAC, audit logs – in days, not months. https://t.co/Xxf2A82ixc

Three things worth considering, as discussed with Martin, in this episode:

1. Multi-region and multi-cloud are risk/cost trade-offs, not best practices.

Martin does not believe that there is a “best practice” in deciding whether to go multi-region or multi-cloud. This decision is a tradeoff between risk and costs. It’s a business decision to be made. Designing Data-Intensive Applications gives engineers the vocabulary to articulate the tradeoffs, not to dictate answers.

2. Replication for fault tolerance is more relevant for most engineers these days than sharding.

Though the book has a full chapter on sharding, Martin said that the cloud has reduced the need for manual sharding for the majority of teams. This is also because machines are increasingly bigger, and more workloads fit on a single machine. Sharding across machines is increasingly a specialist concern; replication for fault tolerance, however, is still relevant at every scale.

3. Knowing system internals as a superpower for application developers.

Martin maintains that Designing Data-Intensive Applications is not a book for people who build databases or even infrastructure, but it’s helpful for application developers to develop an intuition for making good design decisions and debugging performance issues we will eventually encounter.

[Amjad Masad @amasad]

This was so controversial to say last year. Now it’s basically a fact—they’re adding to the Python stdlib.

[Daily Turkic @DailyTurkic]

Someone please start making more films set in early-1900s Istanbul 🇹🇷

[Mustafa Barghouti @Mustafa_Barghouti @MustafaBarghou1]

Israel is the only entity in the world that imposes racist roads segregation, highways for Israelis and bad roads for Palestinians

[Muhammad Smiry 🇵🇸 @MuhammadSmiry]

Gaza today.

[Paul Graham @paulg]

The new plan is apparently to make as many people as possible as miserable as they can be made without blowing them up. And this will be pursued with the same relentless ingenuity as blowing them up was.

[The Linux Foundation @linuxfoundation]

Today we launch the Agentic AI Foundation (AAIF) with project contributions of MCP (@AnthropicAI), goose (@blocks) and https://t.co/KhZ4k71lmO (@OpenAI), creating a shared ecosystem for tools, standards, and community-driven innovation.

Learn more about this major step toward: https://t.co/JHvfl8D0UP

[AIPAC Tracker @TrackAIPAC]

WATCH: Drone footage of southern Gaza released by @AP

This is U.S. taxpayer-funded genocide. This is Zionism.

[Andrej Karpathy @karpathy]

Excited to release new repo: nanochat!
(it's among the most unhinged I've written).

Unlike my earlier similar repo nanoGPT which only covered pretraining, nanochat is a minimal, from scratch, full-stack training/inference pipeline of a simple ChatGPT clone in a single, dependency-minimal codebase. You boot up a cloud GPU box, run a single script and in as little as 4 hours later you can talk to your own LLM in a ChatGPT-like web UI.

It weighs ~8,000 lines of imo quite clean code to:

- Train the tokenizer using a new Rust implementation
- Pretrain a Transformer LLM on FineWeb, evaluate CORE score across a number of metrics
- Midtrain on user-assistant conversations from SmolTalk, multiple choice questions, tool use.
- SFT, evaluate the chat model on world knowledge multiple choice (ARC-E/C, MMLU), math (GSM8K), code (HumanEval)
- RL the model optionally on GSM8K with "GRPO"
- Efficient inference the model in an Engine with KV cache, simple prefill/decode, tool use (Python interpreter in a lightweight sandbox), talk to it over CLI or ChatGPT-like WebUI.
- Write a single markdown report card, summarizing and gamifying the whole thing.

Even for as low as ~$100 in cost (~4 hours on an 8XH100 node), you can train a little ChatGPT clone that you can kind of talk to, and which can write stories/poems, answer simple questions. About ~12 hours surpasses GPT-2 CORE metric. As you further scale up towards ~$1000 (~41.6 hours of training), it quickly becomes a lot more coherent and can solve simple math/code problems and take multiple choice tests. E.g. a depth 30 model trained for 24 hours (this is about equal to FLOPs of GPT-3 Small 125M and 1/1000th of GPT-3) gets into 40s on MMLU and 70s on ARC-Easy, 20s on GSM8K, etc.

My goal is to get the full "strong baseline" stack into one cohesive, minimal, readable, hackable, maximally forkable repo. nanochat will be the capstone project of LLM101n (which is still being developed). I think it also has potential to grow into a research harness, or a benchmark, similar to nanoGPT before it. It is by no means finished, tuned or optimized (actually I think there's likely quite a bit of low-hanging fruit), but I think it's at a place where the overall skeleton is ok enough that it can go up on GitHub where all the parts of it can be improved.

Link to repo and a detailed walkthrough of the nanochat speedrun is in the reply.

[Mohammed Momen @momen_ai]

keep boycotting,
keep isolating.
it’s getting rotten fast.

[Muna Omari منى العمري @MonaOmary]

الحركات العمالية والنقابية في نابولي الإيطالية تنفذ وعدها بإيقاف محطات القطارات والتمهيد لإعلان الإضراب العام في حال تم اعتراض الأسطول، معلومات من هناك تفيد ان المظاهرة أوقفت عمل محطة القطار المركزية ( الظاهرة في الفيديو) في نابولي وتتجه الان الى مبنى المحافظة ..

بينما ترشح الأخبار ان حركة عمال الموانئ في جنوة تتجه في هذه الأثناء لإغلاق الميناء.

الصحافة الإيطالية تتحدث عن اغلاق ميناء ليفورونو وهو من اهم الموانئ الاستراتيجية في ايطاليا، يتعامل الميناء مع نحو 780 ألف حاوية سنويا ويخدم حوالي 2.5 مليون راكب، ويُعد بوابة رئيسية لتجارة السلع بين أوروبا وباقي العالم، خاصة شمال إفريقيا ودول المتوسط، وكان قد رفض استقبال سفن اسرائيلية او محملة بأسلحة ذاهبة لإسرائيل.

يذكر ان النقابات العمالية في إيطاليا دعت لإضراب يوم الجمعة 3.10 إحتجاجا على اعتراض أسطول الصمود لكسر الحصار عن قطاع غزة.

[Amjad Masad @amasad]

Many are asking if they can move their Next.js projects from Vercel to Replit?

Yes!

- Go to Import then GitHub
- enter repo URL
- Agent will take care of the rest

It will set up the dev and deployment environments!

For companies moving lots of work, happy to help + discount.

[Amjad Masad @amasad]

When I first spoke out about the genocide, I was one of the few voices in tech, and it came at a cost.

I faced sabotage especially from the VC class: lies, leaks, threats, and blocked investments.

It was painful, but I never once regretted standing up for the children in Gaza.

Today, the tide in tech has shifted. The truth is undeniable.

If you’ve been holding back, now is the time to speak out and call out anyone supporting or celebrating genocide. It won’t cost you much—in fact, it will earn you respect, and more importantly a clear conscious. Plus, alienating those who will hate you for speaking is a feature, not a bug.

[وكالة أنباء تركيا @tragency1]

وكالة التعاون والتنسيق التركية "تيكا" تدعم المركز الوطني لمكافحة السرطان في عدن اليمنية بمعدات طبية حديثة ما يساعد في تشخيص وعلاج آلاف المرضى ومتابعة حالاتهم بشكل أفضل.

🇹🇷 🇾🇪

[Paul Graham @paulg]

It's official. It's genocide.

[freeCodeCamp.org @freeCodeCamp]

The open source community offers devs many opportunities to collaborate, learn, & grow.

Contributing to OSS helps you develop skills, get key experience when job hunting, and so on.

If you want to dive in, this course will teach you how to get involved.

https://t.co/oVxDLFKrNe

[Paul Graham @paulg]

A Palestinian child's life is worth as much as an Israeli child's life.

[Dear Son. @DearS_o_n]

The taste of peace.