MagedSaeed

Bsc & Msc CS Alumni @KFUPM|#NLP ~ Arabic| AI Engineer & Researcher @ JRCAI, @KFUPM| ex-software engineer @KFUPM IT| Cofounded @arabicml2

[MagedSaeed @_MagedSaeed_]

The hottest new AI company is Anthropic!

https://t.co/11gSpBYUlF

[Andrej Karpathy @karpathy]

Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.

[MagedSaeed @_MagedSaeed_]

صراحة متوقع إن الأداء يتحسن بعد ال fine tuning
الغرض من ال pretraining هو التعرف على هيكل اللغة بغض النظر عن مصدر النص، هيكل اللغة ما تغير تقريبا من قرون.
ال LLM بالنهاية نموذج لمحاكاة اللغات والنصوص، وقدرته على الحفظ والتذكر جاءت كأثر عرضي (side effect) لعملية تدريبه على هذه المحاكاة.

[MagedSaeed @_MagedSaeed_]

Pleased to share Anonymous HF: an open-source service built to solve the pain point of anonymously sharing #HuggingFace artifacts in double blind peer reviews.

The problem: #HuggingFace repos are either fully public (anyone can find them and see who owns them) or fully private (404 for everyone except owners and organization members). There's no built-in way to give visitors anonymous, read-only access to these repos.

With Anonymous HF, authors sign in with their HuggingFace accounts, submit a repo or branch URL, and get a randomized anonymous link. Visitors/Reviewers can browse files, preview content, and download anonymously. Files are streamed directly from HuggingFace in chunks, nothing is stored on this service.

This service is equivalent to Anonymous GitHub (https://t.co/f6DjCIkyN7) but for HuggingFace artifacts. Note that anonymous GitHub gives a storage quota per account (meaning that they stream the repo first to their servers?) while anonymous HF does not.

Key highlights:

- Works with both HF models and datasets.
- Built-in file browser and viewer for reviewers.
- Auto-expiry with manual controls (extend, expire, delete).
- Activity logging for every view and download.
- Optional Colab link so visitors/reviewers can interact with the artifacts.

The site is live on railway: https://t.co/LzVW4jtKYB

Source code: https://t.co/0S8uHIThDx

[MagedSaeed @_MagedSaeed_]

@MRashadnow Well, for the sake of benchmarking. we can start with well documented ones (say between 100 to 500) with different sizes and different languages. We then generate this llm.md (we study different methods of generating it) then compare against their MCPs, what do u think?

[MagedSaeed @_MagedSaeed_]

Actually, following ur line of thoughts, it might be a good idea to have a single file (.llm.md, similar to llm.txt?) inside each package that overviews its purpose and overviews each function stating it's path, desc, inputs, and outputs. Can we benchmark this method against current MCPs?

[MagedSaeed @_MagedSaeed_]

I do agree. Writing such software is not only useful for LLMs, but for the good of all. My point is if we can automate (or at least semi automate) this process of MCP creation, we may not need to invest effort of MCPing currently useful packages yet not well cooked for LLMs. Am I off somewhere?

[MagedSaeed @_MagedSaeed_]

@claudeai I announced this article about claude code today and just another brand new feature just dropped in! u r making it hard to follow up @claudeai

https://t.co/CDCcQe3bb0

[MagedSaeed @_MagedSaeed_]

Sharing what I learned from months of using @AnthropicAI @claudeai and #ClaudeCode in my daily workflow.

I finally finished writing my walkthrough article on agentic AI for research and development on my blog at this link:

https://t.co/krU9Vt48GF

If you want to try Claude Code along with the article, here's a referral link that gives the first 3 people a free week of the subscription:

https://t.co/RzjZy44onZ

What the article tries to cover:

- What actually makes an LLM "agentic", and how does this change how you work?

- Chat vs Agentic LLMs.

- How to set up Claude Code in a project: #CLAUDE.md, the .claude/ directory, and when to use its different modes (edit mode vs. plan mode vs. skip-permissions)?

- Claude Code Plugins and echosystem: hooks, skills, commands, and subagents.

- Finally, and importantly, the limits: hallucinated citations, code hallucination, maintenance burden, cognitive load, and what reproducibility means when an agent is involved.

The article is more of a shared experience. If you have your own setup and workflows, please do share! I would genuinely love to hear how yours differ, and I/we can improve from your experience.

Hope you enjoy walking through.

[MagedSaeed @_MagedSaeed_]

Sharing what I learned from months of using @AnthropicAI @claudeai and #ClaudeCode in my daily workflow.

I finally finished writing my walkthrough article on agentic AI for research and development on my blog at this link:

https://t.co/krU9Vt48GF

If you want to try Claude Code along with the article, here's a referral link that gives the first 3 people a free week of the subscription:

https://t.co/RzjZy44onZ

What the article tries to cover:

- What actually makes an LLM "agentic", and how does this change how you work?

- Chat vs Agentic LLMs.

- How to set up Claude Code in a project: #CLAUDE.md, the .claude/ directory, and when to use its different modes (edit mode vs. plan mode vs. skip-permissions)?

- Claude Code Plugins and echosystem: hooks, skills, commands, and subagents.

- Finally, and importantly, the limits: hallucinated citations, code hallucination, maintenance burden, cognitive load, and what reproducibility means when an agent is involved.

The article is more of a shared experience. If you have your own setup and workflows, please do share! I would genuinely love to hear how yours differ, and I/we can improve from your experience.

Hope you enjoy walking through.

[MagedSaeed @_MagedSaeed_]

@MRashadnow @ahmdelemam أنت موجود برضه هنا يا محمد رشاد،،
😃

[MagedSaeed @_MagedSaeed_]

It is really interesting to see how this LLM stochastic yet powerful tech is converging into a major revolution in software industry.

[MagedSaeed @_MagedSaeed_]

Apparantly, with the current landscape of AI coding agents, it is becoming clear that code is a major market of LLMs vendors. Why? Code:
- can build products & genertae money.
- is well structreded, verbose, redundant -> has less entropy (easy to learn).
- is simple by design & has a closed set of structs (code with variables, conditionals, & loops is a turing-complete code).
- well verifiable as it is merely a translation from high-level business requirements.
- unlike natural languages, cross linguality is easier to transfer.

What can you add to the list?

[MagedSaeed @_MagedSaeed_]

@muotaz5 very well attributed!

[MagedSaeed @_MagedSaeed_]

@Alqahtani_khald طبعا -لفك اللبس- المقصود هنا ال exposure وليس ال replacement:

https://t.co/sFaK9gTmPo

[Andrej Karpathy @karpathy]

@_kaitodev @Ignaci0m_ The "exposure" was scored by an LLM based on how digital the job is. This has no baring on what actually happens to these occupations, which has to do with demand elasticity and a lot more. People are sensationalizing the visualization tool and putting words in my mouth.

[MagedSaeed @_MagedSaeed_]

@MamdouhAI فكرة حلوة، ويش فرق هذا عن إنك تعدل ال memory عنك في claude؟ ولا claude code ما يستطيع يقراها؟

[Mazen — sa/acc @ma7dev]

شاركت مع الحضور تجربتي الشخصية لدعمي لبعض من المشاريع مفتوحة المصدر:
>كيف كنت ادعم المشاريع
>كيف انها ساعدتني في تطوري مهاراتي التقنية والتواصل
>كيف انها خلقت لي فرص وظيفية وذكريات جميلة مع اصحاب المشاريع المفتوحة، بالاخص @zaidalyafeai و @_MagedSaeed_ من مجموعة @arabicml2 ❤️

[MagedSaeed @_MagedSaeed_]

@AIVibecode openrouter ما ينفع في مثل هذي الحالات؟

[MagedSaeed @_MagedSaeed_]

@Adel1989 @KFUPM في تورية في مكان ما لا نعرفه، والمعنى في بطن الشاعر،،
😆

[MagedSaeed @_MagedSaeed_]

سألت gemini pro 3 هذا السؤال:
```
اكتب قصيدة فصيحة بليغة في مدح جامعة الملك
فهد للبترول والمعادن @KFUPM
```
وهذا كان رده:

```
هذه قصيدة فصيحة بليغة، نظمتُها خصيصاً لمدح جامعة الملك فهد للبترول والمعادن (KFUPM)، مستلهماً مكانتها كمنارة للعلم في "جبل الظهران"، ودورها في تخريج قادة الصناعة والتطوير:
​📜 مَنَارَةُ الجَبَل
​عَلَى قِمَمِ الظَّهْرَانِ لِلْمَجْدِ مَوْعِدُ
وَلِلعِلْمِ صَرْحٌ بِالمَكَارِمِ يَصْعَدُ
​تَأَلَّقَ فِي شَرْقِ البِلَادِ كَأَنَّهُ
مَنَارَةُ عِزٍّ فِي السَّمَاءِ تُوَلَّدُ
​إِذَا ذُكِرَتْ "بِتْرُولُ" قَامَتْ هَامَةٌ
تُطَاوِلُ أَعْنَاقَ السَّحَابِ وَتَشْهَدُ
​سَقَاهَا بَنُو عَبْدِ العَزِيزِ عِنَايَةً
فَأَثْمَرَ غَرْسٌ بِالمَعَارِفِ يُسْعَدُ
​تُحِيلُ رِمَالَ البِيدِ تِبْرًا وَعَسْجَدًا
وَتَبْنِي عُقُولًا لِلصِّنَاعَةِ تَرْفِدُ
​سَلِ "البُرْجَ" عَنْهَا حِينَ يُشْرِقُ ضَوْؤُهُ
يُحَدِّثْكَ أَنَّ اللَّيْلَ بِالعِلْمِ يُطْرَدُ
​وَمَعْدِنُهَا طُلَّابُ صِدْقٍ وَهِمَّةٍ
هُمُ النُّخْبَةُ الشَّمَّاءُ وَالجِيلُ الأَوْحَدُ
​تَخَرَّجَ مِنْ سَاحَاتِهَا كُلُّ مُبْدِعٍ
يُشِيدُ بِنَاءَ الدَّارِ فِكْرًا وَيُوجِدُ
​فَيَا جَامِعَ الأَذْكَياءِ دُمْتِ مَنِيعَةً
وَدَامَ لَكِ التَّارِيخُ يَحْكِي وَيَسْرُدُ
​حَمَلْتِ اسْمَ "فَهْدٍ" فَارْتَقَيْتِ مَكَانَةً
يَغَارُ لَهَا النَّجْمُ البَعِيدُ وَيَحْسُدُ
​💎 إضاءات في القصيدة:
​"على قمم الظهران": إشارة لموقع الجامعة الجغرافي المميز على "الجبل".
​"سل البرج": الإشارة إلى برج الجامعة الشهير (Water Tower) الذي يعد أيقونة معمارية ورمزاً لها.
​"تحيل رمال البيد تبراً": كناية عن الاستثمار في الموارد الطبيعية (البترول والمعادن) وتحويل المادة الخام إلى قيمة اقتصادية وعلمية.
​"حملت اسم فهد": الفخر بارتباط اسمها بالملك فهد بن عبدالعزيز رحمه الله.
​هل ترغب في صياغة رسالة رسمية أو كلمة إلقائية تُدرج فيها هذه الأبيات؟
```

ما رأيكم؟؟